├── .gitignore ├── _config.yml ├── abstracts ├── 2022-06-23.md ├── 2019-05-31.md ├── 2020-09-25.md ├── 2019-10-25.md ├── 2021-08-24.md ├── 2019-02-08.md ├── 2020-07-02.md ├── 2019-03-08.md ├── 2019-02-22.md ├── 2019-04-19.md ├── 2021-09-28.md ├── 2019-03-22.md ├── 2019-03-15.md ├── 2019-03-29.md ├── 2021-08-03.md ├── 2019-05-03.md ├── 2019-11-01.md ├── 2021-09-14.md ├── 2020-02-28.md ├── 2022-05-17.md ├── 2021-10-26.md ├── 2021-07-13.md ├── 2019-03-01.md ├── 2020-04-17.md ├── 2019-11-15.md ├── 2020-06-19.md ├── 2020-03-13.md ├── 2019-12-06.md ├── 2020-02-07.md ├── 2022-04-26.md ├── 2021-02-02.md ├── 2019-01-25.md ├── 2019-05-24.md ├── 2020-01-10.md ├── 2021-09-21.md ├── 2020-10-09.md ├── 2020-05-22.md ├── 2022-05-19.md ├── 2022-06-07.md ├── 2019-04-12.md ├── 2019-02-01.md ├── 2020-10-02.md ├── 2022-06-28.md ├── 2020-08-21.md ├── 2020-06-05.md ├── 2022-03-22.md ├── 2019-05-21.md ├── 2020-05-01.md ├── 2021-10-19.md ├── 2021-12-06.md ├── 2022-04-19.md ├── 2020-07-17.md ├── 2021-01-19.md ├── 2022-01-13.md ├── 2021-01-12.md ├── 2020-10-16.md ├── 2020-05-29.md ├── 2021-03-09.md ├── 2022-06-14.md ├── 2020-06-26.md ├── 2020-11-02.md ├── 2021-12-07.md ├── 2020-05-15.md ├── 2021-11-02.md ├── 2020-01-31.md ├── 2021-08-10.md ├── 2020-09-11.md ├── 2020-03-06.md ├── 2021-12-14.md ├── 2022-08-09.md ├── 2019-02-15.md ├── 2020-07-24.md ├── 2021-11-30.md ├── 2022-02-01.md ├── 2020-02-14.md ├── 2021-04-16.md ├── 2020-09-04.md ├── 2022-09-13.md └── 2022-03-15.md ├── 2019.md ├── README.md ├── 2021.md └── 2020.md /.gitignore: -------------------------------------------------------------------------------- 1 | .DS_Store 2 | -------------------------------------------------------------------------------- /_config.yml: -------------------------------------------------------------------------------- 1 | title: NERSC Data Seminars 2 | description: 3 | google_analytics: 4 | theme: jekyll-theme-slate 5 | -------------------------------------------------------------------------------- /abstracts/2022-06-23.md: -------------------------------------------------------------------------------- 1 | # Demo and hands-on session on ReFrame 2 | ## Lisa Gerhardt, Alberto Chiusole - NERSC, Berkeley Lab 3 | 4 | ## Abstract 5 | Overview and brief demo of the capabilities of ReFrame, and how we use it at NERSC to run pipelines on different systems and continuously test the user-facing requirements. 6 | -------------------------------------------------------------------------------- /abstracts/2019-05-31.md: -------------------------------------------------------------------------------- 1 | ## Reflections on Human Space Flight” subtitled “Why Single Planet Species Don’t Survive) 2 | ### James Newman (Space Systems Academic Group (SSAG) at the Naval Postgraduate School) 3 | 4 | During this talk the Speaker reflects on his involvement with the building of the International Space Station and repair of the Hubble Space Telescope and discusses some of the outcomes from these endeavors. Then the talk moves to consider some of the philosophical implications of taking a long-term outlook and the rationale for not being a single-planet species. 5 | -------------------------------------------------------------------------------- /abstracts/2020-09-25.md: -------------------------------------------------------------------------------- 1 | # ExaHDF5: An Update on the ECP HDF5 Project 2 | ## Quincey Koziol (NERSC), Suren Byna (CRD) 3 | 4 | ## Abstract 5 | This talk will present an update on the features and future of HDF5 for exascale HPC. Currently, our work focuses on asynchronous I/O and node-local storage caches, but future work will include GPU direct I/O and data movement across the deeper memory hierarchy anticipated on future systems. 6 | 7 | 8 | ## Bio 9 | Quincey Koziol and Suren Byna lead the LBNL efforts to develop HDF5's capabilities to meet science application I/O needs now and in the future. 10 | -------------------------------------------------------------------------------- /abstracts/2019-10-25.md: -------------------------------------------------------------------------------- 1 | # Opportunities and Challenges in Linking DAQ and HPC Systems 2 | ## David Skinner (Data Science Engagement Group, NERSC) 3 | 4 | This talk looks at HPC from the perspective of data acquisition systems (DAQs) at experimental and observational facilities. A systems view of DOE-scale instruments producing data increasingly includes opportunities and/or needs to modulate computational intensity upwards. Experimental facilities that "plug into HPC" for this purpose often include DAQ hardware and systems for which HPC can be a challenge. This seminar reviews related issues from recent data science engagements at NERSC. 5 | -------------------------------------------------------------------------------- /abstracts/2021-08-24.md: -------------------------------------------------------------------------------- 1 | # JGI Computing - the future is looking cloudy 2 | ## Kjiersten Fagnan (Joint Genome Institute/LBL) 3 | 4 | ## Abstract 5 | The DOE Joint Genome Institute produces high-quality omics data from fungi, plants, microbes, and metagenomes. The computational infrastructure needed to support processing and analysis spans laptops to exascale. JGI has adapted to these needs through moving to a distributed network of computing and storage resources. In this talk I'll describe those resources, what's run where, and the software infrastructure we're building to maintain a high level of usability for JGI's staff and User community. 6 | -------------------------------------------------------------------------------- /abstracts/2019-02-08.md: -------------------------------------------------------------------------------- 1 | ## An Empirical Model of Large-Batch Training 2 | ### Sam McCandlish (OpenAI) 3 | 4 | How quickly can neural networks be trained using large batch sizes? The limits of data parallelism seem differ from domain to domain, ranging from batches of tens of thousands in ImageNet classifiers to batches of millions in RL agents that play the game Dota 2. We describe a simple and easy-to-measure statistic called the gradient noise scale that predicts the largest useful batch size across many applications. Our empirically-motivated theory also describes the tradeoff between compute-efficiency and time-efficiency, and provides a rough model of the benefits of adaptive batch-size training. 5 | -------------------------------------------------------------------------------- /abstracts/2020-07-02.md: -------------------------------------------------------------------------------- 1 | # Workflows at NERSC: Overview and GNU Parallel Parsl, Papermill demos 2 | ## Bill Arndt, Laurie Stephey, Bjoern Enders (NERSC) 3 | 4 | ## Abstract 5 | In fall 2019 NERSC started an effort to reevaluate the workflow tools we aresupporting and exlore new tools. Bill Arndt, Bjoern Enders, and Laurie Stepheywill give an overview of this ongoing effort and will demo 3 tools:GNU-Parallel, Parsl, and Papermill. 6 | 7 | ## Bio 8 | Bill Arndt is a Computer Systems Engineer in the Data Science Engagement Group atNERSC. Bjoern is a Data Science Workflows Architect in the Data ScienceEngagement Group at NERSC. Laurie Stephey is a Data Analytics Engineer in theData and Analytics Services Group at NERSC. 9 | -------------------------------------------------------------------------------- /abstracts/2019-03-08.md: -------------------------------------------------------------------------------- 1 | ## Introduction to Deep Learning 2 | ### Mustafa Mustafa (NERSC, LBL) 3 | 4 | I will introduce the modern incarnation of neural networks, specifically the construction process of deep models that are powering most recent advances in artificial intelligence. Assuming no prior knowledge of the field, the audience will be brought up to speed with how neural networks work, how they are built and optimized and common tricks and techniques to improve their performance. By the end of the talk the audience will be familiar with start-of-the-art convolutional architectures and the stages of developments that brought them about. If time permits, I will overview common problem instances that can be tackled using deep learning. 5 | -------------------------------------------------------------------------------- /abstracts/2019-02-22.md: -------------------------------------------------------------------------------- 1 | ## Learning quantum states with generative models 2 | ### Juan Felipe Carrasquilla (Vector Institute, Toronto, Canada) 3 | 4 | The technological success of machine learning techniques has motivated a research area in the condensed matter physics and quantum information communities, where new tools and conceptual connections between machine learning and many-body physics are rapidly developing. In this talk, I will discuss the use of generative models for learning quantum states. In particular, I will discuss a strategy for learning mixed states through a combination of informationally complete positive-operator valued measures and generative models. In this setting, generative models enable accurate learning of prototypical quantum states of large size directly from measurements mimicking experimental data. 5 | -------------------------------------------------------------------------------- /abstracts/2019-04-19.md: -------------------------------------------------------------------------------- 1 | ## Picture Perfect 2 | ### Peter Denes (LBL) 3 | 4 | Peter Denes, Uli Dahmen and Kenneth Downing were the recipients of the 2015 Berkeley Lab Prize − Lifetime Achievement Award for their scientific advances and leadership in making Berkeley Lab among the world’s forefront centers for electron microscopy. In this talk Peter will describe the big picture in advanced microscopy from detector engineering to data analysis. Detectors in use now can generate images at hundreds of kHz and future designs less constrained by e.g. Moore's Law than CPU/GPU technology. These include instruments such as NCEM's 4DSTEM camera and others in NERSC's super-facility roadmap. This seminar addresses building devices that extract quality data from experimental and observational science as well as the downstream path for data. 5 | -------------------------------------------------------------------------------- /abstracts/2021-09-28.md: -------------------------------------------------------------------------------- 1 | # Fusion Long Range Plan and Fusion Energy Sciences Advisory Committee Report Briefing and Current and Future FES Needs at NERSC 2 | ## Richard Hawryluk (PPPL), Troy Carter (UCLA), Brian Wirth (ORNL), Chris Holland (UCSD), Dave Humphreys (General Atomics) 3 | 4 | ## Abstract 5 | Research in Fusion Energy Sciences comprises a big portion of the NERSC workload. Recently FES has developed a long-range plan to guide research priorities. The leaders of this effort will give a summary of these plans with an emphasis on how they involve NERSC and HPC. Outline: Troy Carter Brief overview of FESAC long-range plan with slide on HPC modeling (15 min), Rich Hawryluk main recommendations from NASEM report and strategic plan (8min), Brian Wirth role integrated design teams and integrated modeling (8min), Dave Humphreys report on FES/ASCR machine learning workshop (15 min Q&A). 6 | -------------------------------------------------------------------------------- /abstracts/2019-03-22.md: -------------------------------------------------------------------------------- 1 | ## Deep Learning For Spatiotemporal Data 2 | ### Rose Yu (Northeastern University) 3 | 4 | Applications such as climate science, intelligent transportation, aerospace control, and sports analytics apply machine learning for large-scale spatiotemporal data. This data is often nonlinear, high-dimensional, and demonstrates complex spatial and temporal correlation. Existing deep learning models cannot handle complex spatiotemporal dependency structures. We'll explain how to design deep learning models to learn from large-scale spatiotemporal data, especially for dealing with non-Euclidean geometry, long-term dependencies, and logical and physical constraints. We'll showcase the application of these models to problems such as long-term forecasting for transportation, long-range trajectories synthesis for sports analytics, and combating ground effect in quadcopter landing for aerospace control. 5 | -------------------------------------------------------------------------------- /abstracts/2019-03-15.md: -------------------------------------------------------------------------------- 1 | ## SENSE: SDN for End-to-end Networked Science at the Exascale 2 | ### Chin Guok (ESNet, LBL) 3 | 4 | Today, domain science applications and workflow processes are forced to view the network as an opaque infrastructure into which they inject data and hope that it emerges at the destination with an acceptable Quality of Experience. There is little ability for applications to interact with network to exchange information, negotiate performance parameters, discover expected performance metrics, or receive status/troubleshooting information in real time. As a result, domain science applications frequently suffer poor performance, especially so in highly distributed environments. This talk will present a vision, and an early instantiation, of an Intelligent Network Service Plane, where the application workflow agents can engage, interact, and obtain custom services from a smart networked ecosystem. 5 | -------------------------------------------------------------------------------- /abstracts/2019-03-29.md: -------------------------------------------------------------------------------- 1 | ## Bridging length scales in granular systems using machine learning 2 | ### Utkarsh Mital (CalTech) 3 | 4 | In this talk, I draw an analogy between images and granular systems. This enables the repurposing of computer vision algorithms to bridge microscopic and macroscopic length scales and model the mechanics of granular materials. I show that convolutional neural networks can be trained to learn the mechanics of grain-scale interactions, enabling them to model high-level macroscopic properties such as stress. I also demonstrate how GANs could be used to synthetically generate grain-scale data that mimics the distribution of real data. Such a strategy could be further extended to bridge site-specific and regional scales, and employed in the context of downscaling satellite datasets. I discuss an example application of mapping regional liquefaction hazards that could benefit by such advances. 5 | -------------------------------------------------------------------------------- /abstracts/2021-08-03.md: -------------------------------------------------------------------------------- 1 | # Legate: High Productivity High Performance Computing 2 | ## Manolis Papadakis (NVIDIA) 3 | 4 | 5 | ## Abstract 6 | The Legate project (https://github.com/nv-legate) aims to provide distributed and accelerated drop-in replacements for popular scientific / data-science libraries (NumPy and Pandas so far). Our goal is to allow programmers to prototype their applications on their local machine, then be able to transparently scale up to large clusters, utilizing the available acceleration hardware, without having to rewrite their application in an explicit parallel programming model like MPI or Dask. By building all Legate libraries on top of Legion's distributed data model and runtime we can achieve seamless asynchronous execution across libraries, with minimal blocking and copying. 7 | 8 | ## Bio 9 | Manolis Papadakis is a Senior Software Engineer at NVIDIA, working on the Legate project. 10 | -------------------------------------------------------------------------------- /abstracts/2019-05-03.md: -------------------------------------------------------------------------------- 1 | ## Infusing Structure into Machine Learning Algorithms 2 | ### Animashree Anandkumar (Caltech) 3 | 4 | Standard deep-learning algorithms are based on a function-fitting approach that do not exploit any domain knowledge or constraints. This makes them unsuitable in applications that have limited data or require safety or stability guarantees, such as robotics. By infusing structure and physics into deep-learning algorithms, we can overcome these limitations. There are several ways to do this. For instance, we use tensorized neural networks to encode multidimensional data and higher-order correlations. We combine symbolic expressions with numerical data to learn a domain of functions and obtain strong generalization. We combine baseline controllers with learnt residual dynamics to improve landing of quadrotor drones. These instances demonstrate that building structure into ML algorithms can lead to significant gains. 5 | -------------------------------------------------------------------------------- /abstracts/2019-11-01.md: -------------------------------------------------------------------------------- 1 | ## Machine Learning, Synthetic Biology and Automation: Engineering Life for the Benefit of Society 2 | ### Hector Garcia Martin (LBNL, JBEI) 3 | 4 | Biology has changed radically in the last two decades, transitioning from a descriptive science into a design science. Synthetic biology leverages engineering approaches to produce biological systems to a given specification (e.g. produce x grams of a medical drug or invade this type of cancer cell). In this effort, new tools are now available that promise to disrupt this field: from CRISPR-enabled genetic editing, to high-throughput omics phenotyping, and exponentially growing DNA synthesis capabilities. However, our inability to predict the behavior of biological systems hampers synthetic biology from reaching its full potential. 5 | 6 | We will show how the combination of machine learning and automation enables the creation of a predictive synthetic biology for the benefit of society. 7 | -------------------------------------------------------------------------------- /abstracts/2021-09-14.md: -------------------------------------------------------------------------------- 1 | # Challenges and successes with a hybrid multicloud implementation for research computing 2 | ## Jonathan Skone (NERSC/LBL) 3 | 4 | ## Abstract 5 | A review of the efforts to evolve an academic on-premises HPC ecosystem to the cloud through the hybrid multicloud solution named Skyway, will be presented. Skyway incorporates multicloud computing resources as elastic extensions of its on-premises HPC infrastructure and makes use of a system-level software package to interface the job scheduler and cloud SDKs, resulting in a seamless experience for users when interacting with both on-premises and cloud resources. The implementation is general enough to interface with one or more cloud providers, which currently include both Amazon AWS and Google GCP. The challenges encountered and the use cases where it has been successful, will be elaborated. 6 | 7 | ## Bio 8 | https://www.nersc.gov/about/nersc-staff/advanced-technologies-group/jonathan-skone/ 9 | -------------------------------------------------------------------------------- /abstracts/2020-02-28.md: -------------------------------------------------------------------------------- 1 | # Time-series Analysis of ESnet Network Traffic: Statistical and Deep Learning models 2 | ## The Superfacility Project Team 3 | 4 | ## Abstract 5 | The Superfacility Initiative was a key component of the CS Area strategic report, and described the research and engineering required to connect experimental, networking and HPC facilities to accelerate scientific discovery. The Superfacility project was created in early 2019 in response to this Initiative. The project tracks, coordinates and communicates the work being performed across the CS Area to address the needs described in the Strategic Plan. This includes close partnership with several science teams whose needs are driving our work. In this talk, we will introduce the Superfacility concept and project structure, and the project leads will present their highlight achievements in 2019, and plans for 2020. 6 | 7 | ## Bio 8 | The Superfacility project includes staff from NERSC, ESnet and CRD. 9 | -------------------------------------------------------------------------------- /abstracts/2022-05-17.md: -------------------------------------------------------------------------------- 1 | # Memory Disaggregation: Potentials and Pitfalls 2 | ## Nan Ding (Computer Science Department, Lawrence Berkeley National Laboratory) 3 | 4 | ## Abstract 5 | Memory usage imbalance has been consistently observed in many data centers. This has sparked interest in memory disaggregation, which allows applications to use all available memory across an entire data center instead of being confined to the memory of a single server. In the talk, I'll present the design space and implementation for building a disaggregated memory system. I'll then discuss the critical metrics for applications to benefit from memory disaggregation. 6 | 7 | ## Bio 8 | Nan Ding is a Research Scientist in the Performance and Algorithms group of the Computer Science Department at Lawrence Berkeley National Laboratory. Her research interests include high-performance computing, performance modeling, and auto-tuning. Nan received her Ph.D. in computer science from Tsinghua University, Beijing, China in 2018. 9 | -------------------------------------------------------------------------------- /abstracts/2021-10-26.md: -------------------------------------------------------------------------------- 1 | # Challenges and Directions in ML System Performance: The MLPerf Story 2 | ## David Kanter (MLCommons) 3 | 4 | ## Abstract 5 | As the industry drives towards more capable ML, workloads are rapidly evolving and the need for performance is nearly unlimited. We explore the challenges and design choices behind MLPerf, the industry standard benchmark for ML system performance. 6 | 7 | ## Bio 8 | David Kanter is a Founder and the Executive Director of MLCommons™ where he helps lead the MLPerf™ benchmarks and other initiatives. He has 16+ years of experience in semiconductors, computing, and machine learning. He founded a microprocessor and compiler startup, was an early employee at Aster Data Systems, and has consulted for industry leaders such as Intel, Nvidia, KLA, Applied Materials, Qualcomm, Microsoft and many others. David holds a Bachelor of Science degree with honors in Mathematics with a specialization in Computer Science, and a Bachelor of Arts with honors in Economics from the University of Chicago. 9 | -------------------------------------------------------------------------------- /abstracts/2021-07-13.md: -------------------------------------------------------------------------------- 1 | # Monitoring Scientific Python Usage at NERSC 2 | ## Rollin Thomas (NERSC/LBL) 3 | 4 | ## Abstract 5 | Last year about 30% of all NERSC users ran jobs involving Python in some way. The most 6 | popular Python package in use in Python jobs isn't NumPy, it's multiprocessing. And among 7 | Python-based MPI jobs, AstroPy appears to be the most significant package in use. How do we 8 | know this and how does this information help us? In this talk, I will discuss how NERSC 9 | monitors the use of Python on its systems, as part of its Monitoring of Data Services (MODS) 10 | project. The talk will cover how the data is collected, stored, analyzed, and published for 11 | consumption by various stakeholders (staff, management, developers, etc.) using a Jupyter 12 | notebook-centric workflow involving GPUs, Dask, Papermill, Voilà, and Spin. 13 | 14 | ## Bio 15 | Rollin Thomas is a Data Architect in the Data and Analytics Services group at NERSC. From 16 | 2015 to 2020 he was in charge of Python support on Cori. 17 | -------------------------------------------------------------------------------- /abstracts/2019-03-01.md: -------------------------------------------------------------------------------- 1 | ## Jupyter at NERSC 2 | ### Rollin Thomas (NERSC, LBL) 3 | 4 | Scientists love using Jupyter because it combines text, visualization, data analytics, and code into a document they can share, modify, or even publish. Many of Cori's data-friendly design features come together through Jupyter to provide a new interface to NERSC resources for doing science. The Spin container-as-a-service platform provides a robust layer for managing the web front-end, JupyterHub. Jupyter plays an important part in NERSC's Superfacility project. On average, 100 people per day use Jupyter "notebooks" through shared login nodes on Cori reserved and configured for the purpose. These notebooks can even communicate with jobs running on Cori compute nodes. Software-defined networking will soon allow users to run notebooks on compute nodes themselves. In this seminar we'll discuss the history of Jupyter at NERSC, get into the nuts and bolts of how Jupyter and JupyterHub work at NERSC, and talk about work we're doing now to enrich the Jupyter in HPC experience. Warning: Live demo. 5 | -------------------------------------------------------------------------------- /abstracts/2020-04-17.md: -------------------------------------------------------------------------------- 1 | # A Data-Driven Global Weather Model Using Reservoir Computing 2 | ## Troy Arcomano (Texas A&M University in the Atmospheric Sciences Department) 3 | 4 | ## Abstract 5 | Data-driven approaches to predict chaotic spatiotemporal dynamical systems have been shown to be successful for a number of high-dimensional, complex systems. One of the most important chaotic systems which impacts our lives is the atmosphere. This, naturally, leads to the question whether a purely data-driven machine learning algorithm can accurately predict the weather. In this talk, we present a prototype machine learning model that can skillfully predict the three dimensional state of the atmosphere for 3-5 days. The training of the machine learning model is computationally efficient and parallelized over thousands of computer cores. Our results suggest that machine learning has the potential to improve the prediction of atmospheric state variables most affected by parameterized processes in numerical models. 6 | 7 | References: Arcomano et al. "A Machine-Learning-Based Global Atmospheric Forecast Model." (2020). https://www.essoar.org/doi/pdf/10.1002/essoar.10502527.1 8 | -------------------------------------------------------------------------------- /abstracts/2019-11-15.md: -------------------------------------------------------------------------------- 1 | # FlowPM: Particle-Mesh N-body Simulation in TensorFlow 2 | ## Chirag Modi (UC Berkeley) 3 | 4 | The upcoming generation of cosmological surveys such as DESI or LSST will probe the Universe on an unprecedented scale and with unparalleled precision, to answer fundamental questions about Dark Matter and Dark Energy. However, optimally extracting cosmological information from this massive amount of data remains a major challenge, and constitutes a very active research area. Having access to differentiable forward simulations of these surveys paves the way to novel and extremely powerful gradient-based inference techniques. For instance, we have demonstrated potential for over a 50% information gain in constraining Dark Energy using the upcoming DESI galaxy survey. In this talk, we will present FlowPM, the first differentiable cosmological N-body simulation code implemented in TensorFlow for seamless integration with deep learning components and gradient-based inference techniques. After showcasing a few examples of the benefits of such a tool, we will discuss our efforts to scale these simulations to large supercomputers using the Mesh TensorFlow framework. 5 | -------------------------------------------------------------------------------- /abstracts/2020-06-19.md: -------------------------------------------------------------------------------- 1 | # Status of Containers in HPC 2 | ## Shane Canon, Data & Analytics Group, NERSC 3 | 4 | ## Abstract 5 | Containers have quickly gained traction in HPC and Data Intensive computing. Containers provides users with greater flexibility, enables sharing and reproducibly, can make workflows more portable and can even improve performance. In this talk we will review some of these benefits, the status of containers at NERSC, and trends for containers in HPC. We will also discuss some of the use cases and success stories for containers at NERSC. 6 | 7 | ## Bio 8 | Shane Canon is a Senior Engineer at Lawrence Berkeley Lab where he works in the NERSC Supercomputing Facility. Over Shane's 20 year career, he has focused on enabling scientists to conduct breakthrough science using large-scale systems including some of the fastest computers and storage systems in the world. Most recently Shane has focused on enabling data-intensive computing and container computing on Supercomputing systems. Shane is also a senior member of the DOE KnowledgeBase (KBase) project and the National Microbiome Data Collaborative which are two projects focused on advancing genomics science. 9 | -------------------------------------------------------------------------------- /abstracts/2020-03-13.md: -------------------------------------------------------------------------------- 1 | # ECP HDF5 - New features and applications 2 | ## Suren Byna (CRD, LBL), Quincey Koziol (NERSC, LBL) 3 | 4 | ## Abstract 5 | HDF5 is a data model, file format, and I/O library that has become a de facto standard for HPC applications to achieve scalable I/O and for storing and managing big data from computer modeling, large physics experiments and observations. Several Exascale Computing Project (ECP) applications are currently using or planning to use HDF5 for I/O. The ExaHDF5 project team of the ECP is working on developing and productizing various features to improve the efficiency of parallel I/O to take advantage of exascale architectures. In this presentation, we will talk about these features, including Virtual Object Layer (VOL), asynchronous I/O, subfiling, Data Elevator, independent metadata updates, querying, etc. The presentation also include integration of HDF5 into ECP applications and co-design efforts, such as EQSIM and AMReX. 6 | 7 | ## Bio 8 | Quincey Koziol is a Principal Data Architect at Berkeley Lab. At NERSC, he drives scientific data architecture discussions, participates in NERSC system storage and I/O design, and continues work on HDF5. 9 | -------------------------------------------------------------------------------- /abstracts/2019-12-06.md: -------------------------------------------------------------------------------- 1 | # Natural Language Processing for Materials Discovery and Design 2 | ## John Dagdelen (LBNL, UC Berkeley) 3 | 4 | The majority of all materials data is currently scattered across the text, tables, and figures of millions of scientific publications. In my talk, I will present the work of our team at Lawrence Berkeley National Laboratory on the use of natural language processing (NLP) and machine learning techniques to extract and discover materials knowledge through textual analysis of the abstracts of several million journal articles. With this data we are exploring new avenues for materials discovery and design, such as how functional materials like thermoelectrics can be identified by using only unsupervised word embeddings for materials. To date, we have used advanced techniques for named entity recognition to extract more than 100 million mentions of materials, structures, properties, applications, synthesis methods, and characterization techniques from our database of over 3 million materials science abstracts. With this data, we are developing machine learning tools for autonomously building databases of materials-properties data extracted from unstructured materials text. 5 | -------------------------------------------------------------------------------- /abstracts/2020-02-07.md: -------------------------------------------------------------------------------- 1 | # Time-series Analysis of ESnet Network Traffic: Statistical and Deep Learning models 2 | ## Mariam Kiran (ESnet, CRD, LBNL) 3 | 4 | ## Abstract 5 | Predicting network traffic can provide information on large data movement and details on how users interact with the network. As the area of ML matures, we are investigating building predictive models that can provide suitable predictions into the future on how traffic will behave on the network links. The goals of this prediction is to investigate anomaly detection and understanding congestion patterns on the network to help manage them more efficiently. In this talk, we will present our results on time series analysis and also show we plan to deploy these models to perform real-time ML predictions. 6 | 7 | ## Bio 8 | Mariam Kiran is a research scientist with shared positions with Energy Sciences Network and the Scientific Data Management (SDM) group in Computational Research Division. Her work specifically concentrates on using advanced software and machine learning techniques to advance system architectures, particularly high-speed networks such as DOE networks. Kiran is the recipient of the DOE ASCR Early Career Award in 2017. 9 | -------------------------------------------------------------------------------- /abstracts/2022-04-26.md: -------------------------------------------------------------------------------- 1 | # FirecREST, RESTful HPC 2 | ## Juan Pablo Dorsch (CSCS) 3 | 4 | ## Abstract 5 | FirecREST is a RESTful API to HPC that empowers scientific communities to access compute and data HPC services and infrastructure through a web interface. This API supports and enhances the development of scientific portals that allow web developers and HPC users to adapt their workflows in a more flexible, secure, automated, and standardized way. In this talk, we will present FirecREST and provide an introduction to its capabilities. 6 | 7 | ## Bio 8 | Juan Pablo Dorsch is a software engineer and lead for the Innovative Resource Access Methods at the CSCS Swiss National Supercomputing Centre. His areas of expertise include microservice architecture design, IAM, web development and RESTful services. Before joining CSCS, Juan held the position of HPC engineer with the Computational Methods Research Centre (CIMEC), and the position of scientific software developer with the International Centre for Numerical Methods in Engineering (CIMNE). He was also previously a degree professor at the National University (UNL) of Littoral in Santa Fe, Argentina. He holds a degree in Informatics Engineering with an emphasis on scientific applications from UNL. 9 | -------------------------------------------------------------------------------- /abstracts/2021-02-02.md: -------------------------------------------------------------------------------- 1 | # Machine learning as a tool for Standard Model measurements 2 | ## Vinicius Mikuni (University of Zurich) 3 | 4 | ## Abstract 5 | The Standard Model of particle physics predicts the interactions between the elementary particles that compose our universe. To study these interactions, protons collide at very high energies at the Large Hadron Collider, generating hundreds of particles through every bunch crossing. In this talk, I will present two measurements: the determination of the ttbb and bbH(bb) cross sections. I will show the difficulties of these determinations and how modern machine learning methods are used to enhance the precision of these measurements. 6 | 7 | ## Bio 8 | I'm originally from Brazil and finished a Bachelor's degree in physics in 2015 at the University of Sao Paulo. At the same university, I did a Masters in astroparticle physics as a member of the AMS Collaboration, measuring the time variation of electron and positron cosmic ray fluxes. In 2017, I became a PhD candidate at the University of Zurich, working on the field of collider physics as a member of the CMS Collaboration. There I work on precision measurements of Standard Model parameters and how machine learning can be applied to improve measurement precision. 9 | -------------------------------------------------------------------------------- /abstracts/2019-01-25.md: -------------------------------------------------------------------------------- 1 | ## Hierarchical Deep Learning for Long-term Sequence Generation 2 | ### Stephan Zheng (Salesforce Research) 3 | 4 | Deep learning is a powerful framework for representation learning, and has proven itself in computer vision, natural language processing and other domains. However, it is still challenging to apply deep learning to structured prediction tasks on spatiotemporal data, e.g., multi-agent human tracking data and physical systems. Here, a key challenge is long-term sequence generation: given an initial state, how can we extrapolate into the far future when the underlying dynamics is nonlinear? 5 | 6 | In this talk, I will first present a family of hierarchical deep learning methods that significantly improves on this task, including on data from real-life sports tracking and synthetic dynamical systems. These include single and multi-agent hierarchical neural networks that both predict long-term goals and short-term actions, learned using auxiliary goal labels. I will then show how multi-resolution non-autoregressive sequence models can learn to forecast by joint extrapolation and interpolation in a completely unsupervised way. Finally, I will present Tensor-Train RNNs, which can efficiently learn long-term forecasting over nonlinear dynamics. 7 | -------------------------------------------------------------------------------- /abstracts/2019-05-24.md: -------------------------------------------------------------------------------- 1 | ## Maglev and the Future of Long Distance Transportation 2 | ### John Van Rosendale (Director Computational Science (retired), College of William and Mary) 3 | 4 | Maglev is faster and quieter than high-speed rail and has lower life-cycle cost. The challenge is the large amount of infrastructure required—maglev is completely incompatible with the vast existing rail infrastructure. With an estimated 1,370,000 km of rail right-of-way worldwide this is a real issue, especially in urban areas where obtaining new right-of-way is very difficult and expensive. 5 | 6 | This talk will present a new maglev design, known as “MagFlite,” that uses highly-efficient permanent-magnet-based levitation. One version of MagFlite is designed to be intercompatible with conventional rail, allowing trains to transition at speed between railroad track and maglev guideways—an idea that can greatly simplifying introduction of maglev around the world. The talk will end with a discussion of futuristic high-vacuum maglev. Continent-scale high-vacuum maglev would be far more convenient and efficient than airline travel, reducing typical travel times by a factor of 10. Simulation of a hypothetical high-vacuum maglev interconnecting the 100 largest population centers in the continental U.S. is currently in progress. 7 | -------------------------------------------------------------------------------- /abstracts/2020-01-10.md: -------------------------------------------------------------------------------- 1 | # Independent metadata updating for large scale parallel I/O systems 2 | ## Tonglin Li (NERSC, LBNL) 3 | 4 | ## Abstract 5 | For bridging the ever-widening performance gap between computation and storage systems, new tiers are introduced to the already deep storage hierarchy. I/O middleware, such as HDF5, has been developed and used for decades to provide applications relatively simple APIs and hide all the low-level details of the underlying I/O and storage systems. As the systems scale out, some of the old designs such as the collective metadata updating mechanism that fits smaller scales start to show the performance penalty. For addressing this problem, we have built a new HDF5-based I/O middleware prototype that enables independent metadata updating. In this presentation, I'll talk about the challenges, the design and implementation of the solutions, and the lessons we learned from building the system. 6 | 7 | ## Bio 8 | Tonglin Li is a computer system engineer at NERSC, Lawrence Berkeley National Laboratory. Before joining LBL, he was a postdoctoral researcher at Oak Ridge National Laboratory. He received his Ph.D. degree in Computer Science from Illinois Institute of Technology in 2015. His research interests include distributed systems, large scale storage systems, cloud computing, high-performance computing, and big data systems. 9 | -------------------------------------------------------------------------------- /abstracts/2021-09-21.md: -------------------------------------------------------------------------------- 1 | # Darshan 3.3.1 & Autoperf 2.0 updates 2 | ## Sudheer Chunduri (Argonne National Lab), Kevin Harms (Argonne National Lab) 3 | 4 | ## Abstract 5 | AutoPerf is a lightweight profiling tool used for Automatic performance collection of MPI (focused on MPI 2.0 operations) usage and hardware performance counter information. AutoPerf 1.0 was deployed on Argonne’s earlier general machine called Mira and was successfully collected logs for over 4-5 years. The analysis from this data was helpful in providing several insights in the MPI space* and beyond. Considering the feedback from this study and to enhance the coverage for more MPI 3.0 operations and other improvements in the recorded and reported summary data, Autoperf 2.0 is designed. AutoPerf2.0 implements two additional Darshan instrumentation modules that can provide details on application MPI communication usage and application performance characteristics on Cray XC platforms. We will describe our plans for future work in this topic and provide a summary of Darshan latest release. 6 | 7 | ## Bio 8 | Sudheer Chunduri is a member of Performance Engineering team at ALCF working on the interconnects and MPI developments and performance analysis. 9 | 10 | Kevin Harms is a Performance Engineering Team Lead at ALCF working on Parallel IO, storage and platform analysis and benchmarking aspects. 11 | -------------------------------------------------------------------------------- /abstracts/2020-10-09.md: -------------------------------------------------------------------------------- 1 | # Generative neural networks: Data-driven simulations for particle physics 2 | ## Ramon Winterhalder (Heidelberg University) 3 | 4 | ## Abstract 5 | First-principle simulations are a key ingredient in LHC analyses. They are crucial to connect fundamental theories with observable quantities. However, state-of-the-art methods are limited by the high-dimensionality of the phase space, the complexity of the simulation task, and non-trivial correlations. Some of these problems can be alleviated when we add generative neural networks to our toolbox. I will demonstrate how generative adversarial networks (GANs) can be used as a data-driven simulation technique and how technical challenges like training instabilities can be tackled or circumvented. In addition, GANs can be modified to either generate new samples following the multi-dimensional difference of distributions or to accommodate weighted input data while still producing unweighted events. Finally, we can invert the full simulation chain with conditional invertible neural networks (cINNs) and unfold high-dimensional distributions in a statistically meaningful manner. 6 | 7 | 8 | ## Bio 9 | Ramon Winterhalder is a PhD student in Tilman Plehn’s research group at the Heidelberg University and works on particle phenomenology at the LHC, applying machine learning methods to supplement current simulation techniques. 10 | -------------------------------------------------------------------------------- /abstracts/2020-05-22.md: -------------------------------------------------------------------------------- 1 | # Learned discretizations for passive scalar advection in a 2-D turbulent flow 2 | ## Jiawei Zhuang (Harvard University) 3 | 4 | ## Abstract 5 | The computational cost of fluid simulations increases rapidly with grid resolution. This has given a hard limit on the ability of simulations to accurately resolve small scale features of complex flows. Here we use a machine learning approach to learn a numerical discretization that retains high accuracy even when the solution is under-resolved with classical methods. We apply this approach to passive scalar advection in a two-dimensional turbulent flow. The method maintains the same accuracy as traditional high-order flux-limited advection solvers, while using 4× lower grid resolution in each dimension. The machine learning component is tightly integrated with traditional finite-volume schemes and can be trained via an end-to-end differentiable programming framework. The solver can achieve near-peak hardware utilization on CPUs and accelerators via convolutional filters. Code is available at https://github.com/google-research/data-driven-pdes (preprint: https://arxiv.org/abs/2004.05477). 6 | 7 | ## Bios 8 | Jiawei Zhuang is a 4-th year PhD at Harvard with a research focus on Earth system modeling and high-performance computing. He has interned at Google Research working on machine-learning-enhanced physical simulations. 9 | 10 | 11 | 12 | -------------------------------------------------------------------------------- /abstracts/2022-05-19.md: -------------------------------------------------------------------------------- 1 | # Quantum Computing for NERSC Staff 2 | ## Katie Klymko, Daan Camps & Jan Balewski (NERSC, Lawrence Berkeley National Laboratory) 3 | 4 | ## Abstract 5 | This will be an introduction to quantum computing discussion with a focus on gate-based quantum algorithms. We will cover topics (at a high level) such as what is a qubit, what are quantum gates, and how to read quantum circuit diagrams. We will end with a discussion of common quantum algorithms, with a deep dive into one popular hybrid quantum-classical algorithm, the variational quantum eigensolver. 6 | 7 | ## Bios 8 | Katie Klymko received her PhD in 2018 from UC Berkeley where she worked on the statistical mechanics of non-equilibrium systems using computational and analytical techniques. She was a postdoc at LBNL from October of 2018 through September of 2021, working on a range of topics including fluctuating hydrodynamics/finite volume methods for modeling mesoscale systems and more recently quantum computing algorithms. Her work in quantum computing has focused on the development of efficient methods for eigenvalue calculations in molecular systems as well as quantum computing algorithms to explore thermodynamic properties. In October of 2021, she became a staff member at NERSC where she is working to integrate HPC and quantum computing. 9 | 10 | Daan Camps is a staff member in the Advanced Technology Group at NERSC. His work focuses on integrating emerging quantum technologies in future generation HPC systems. 11 | -------------------------------------------------------------------------------- /abstracts/2022-06-07.md: -------------------------------------------------------------------------------- 1 | # Building a Platform for Operating Multi-Institutional Distributed Services 2 | ## Lincoln Bryant (Enrico Fermi Institute - University of Chicago) 3 | 4 | ## Abstract 5 | Much of science today is propelled by research collaborations that require highly interconnected instrumentation, computational, and storage resources that cross institutional boundaries. To provide a generalized service infrastructure for multi-institutional science, we propose a new abstraction and implementation of this model: Federated Operations (FedOps) and SLATE. We will show the general principles behind the FedOps trust model and how the SLATE platform implements FedOps for building a service fabric over independently operated Kubernetes clusters. Finally, we will show how SLATE is being used to manage data and software caching networks in production across computing sites in the US ATLAS computing facility in support the ATLAS experiment at the CERN Large Hadron Collider. 6 | 7 | ## Bio 8 | Lincoln Bryant is a Research Engineer in the Enrico Fermi Institute at the University of Chicago. He has over a decade of experience building and supporting High-Throughput Computing (HTC), distributed storage, and containerization/virtualization systems for both the ATLAS experiment at the Large Hadron Collider and other collaborations as part of the Open Science Grid Consortium. Lincoln is one of the primary contributors to the Services Layer At The Edge (SLATE) project and has been an active Kubernetes user since 2017. 9 | -------------------------------------------------------------------------------- /abstracts/2019-04-12.md: -------------------------------------------------------------------------------- 1 | ## Sizing Neural Network Experiments 2 | ### Gerald Friedland (UC Berkeley and LLNL) 3 | 4 | Most contemporary machine learning experiments are performed treating the underlying algorithms as a black box. This approach, however, fails when trying to budget large scale experiments or when machine learning is used as part of scientific discovery and uncertainty needs to be quantifiable. Using the example of Neural Networks, this talk presents a line of research enabling the measurement and prediction of the capabilities of machine learners, allowing a more rigorous experimental design process for machine learning experiments. The main idea is taking the viewpoint that memorization is worst-case generalization. My presentation is made of three parts. Based on MacKay's information theoretic model of supervised machine learning~\cite{mackay2003}, I first derive four easily applicable engineering principles to analytically determine the upper-limit capacity of neural network architectures. This allows the comparison of the efficiency of different architectures independent of a task. Second, I introduce and experimentally validate a heuristic method to estimate the neural network capacity requirement for a given learning task. Third, I outline a generalization process that successively reduces the capacity starting at the memorization estimate. I conclude with a discussion on the consequences of sizing a machine learner wrongly, which includes a potentially increased number of adversarial examples. 5 | -------------------------------------------------------------------------------- /abstracts/2019-02-01.md: -------------------------------------------------------------------------------- 1 | ## Mesh-TensorFlow: Deep Learning for Supercomputers 2 | ### Noam Shazeer (Google Brain) 3 | 4 | Batch-splitting (data-parallelism) is the dominant distributed Deep Neural Network (DNN) training strategy, due to its universal applicability and its amenability to Single-Program-Multiple-Data (SPMD) programming. However, batch-splitting suffers from problems including the inability to train very large models (due to memory constraints), high latency, and inefficiency at small batch sizes. All of these can be solved by more general distribution strategies (model-parallelism). Unfortunately, efficient model-parallel algorithms tend to be complicated to discover, describe, and to implement, particularly on large clusters. We introduce Mesh-TensorFlow, a language for specifying a general class of distributed tensor computations. Where data-parallelism can be viewed as splitting tensors and operations along the "batch" dimension, in Mesh-TensorFlow, the user can specify any tensor-dimensions to be split across any dimensions of a multi-dimensional mesh of processors. A Mesh-TensorFlow graph compiles into a SPMD program consisting of parallel operations coupled with collective communication primitives such as Allreduce. We use Mesh-TensorFlow to implement an efficient data-parallel, model-parallel version of the Transformer sequence-to-sequence model. Using TPU meshes of up to 512 cores, we train Transformer models with up to 5 billion parameters, surpassing state of the art results on WMT’14 English-to-French translation task and the one-billion-word language modeling benchmark. Mesh-Tensorflow is available at https://github.com/tensorflow/mesh 5 | -------------------------------------------------------------------------------- /abstracts/2020-10-02.md: -------------------------------------------------------------------------------- 1 | # Enabling Interactive, On-Demand High Performance Computing for Rapid Prototyping and Machine Learning 2 | ## Albert Reuther (MITLincoln Laboratory Supercomputing Center) 3 | 4 | 5 | ## Abstract 6 | For decades, the use of HPC systems was limited to those in the physical scienceswho had mastered their domain in conjunction with a deep understanding of HPCarchitectures and algorithms. During these same decades, consumer computingdevice advances produced laptops, tablets, and smartphones that allow millionsto interactively develop and share code projects using high productivitylanguages and environments. The HPC community faces many challenges associatedwith guiding researchers from disciplines that routinely utilize highproductivity interactive tools to effectively use HPC systems, since it is fruitlessto expect them to give up the interactive, on-demand nature of their workflows. 7 | 8 | For over adecade, MIT Lincoln Laboratory has been supporting interactive, on-demand HPCby seamlessly integrating familiar high productivity tools to provide userswith an increased number of design turns, rapid prototyping capability, andfaster time to insight. In this talk, we discuss the lessons learned whilesupporting interactive, on-demand high performance computing from theperspectives of the users and the team supporting the users and the system. Atits core, it involves an expansion of what the HPC ecosystem paradigm entailsincluding expansions in system architecture, scheduler policies, metrics ofsuccess, and supported software development environments and tools. We concludewith how our team supports users and the systems in this paradigm expansion. 9 | -------------------------------------------------------------------------------- /abstracts/2022-06-28.md: -------------------------------------------------------------------------------- 1 | # FourCastNet: Data-driven, high-resolution atmosphere modeling at scale 2 | ## Shashank Subramanian, Data & Analytics Services Group, National Energy Research Scientific Computing Center (NERSC), Lawrence Berkeley National Laboratory 3 | 4 | ## Abstract 5 | We present FourCastNet, short for Fourier Forecasting Neural Network, a global data-driven weather forecasting model that provides accurate short to medium-range global predictions at 25km resolution. FourCastNet accurately forecasts high-resolution, fast-timescale variables such as the surface wind speed, total precipitation, and atmospheric water vapor with important implications for wind energy resource planning, predicting extreme weather events such as tropical cyclones and atmospheric rivers, as well as extreme precipitation. We compare the forecast skill of FourCastNet with archived operational IFS model forecasts and find that the forecast skill of our purely data-driven model is remarkably close to that of the IFS model for short to medium-range forecasts. FourCastNet generates a week-long forecast in less than 2 seconds, orders of magnitude faster than IFS, enabling the creation of inexpensive large-ensemble forecasts for improved probabilistic forecasting. Finally, our implementation is optimized and we present efficient scaling results on different supercomputing systems up to 3808 NVIDIA A100 GPUs, resulting in 80000 times faster time-to-solution relative to IFS, in inference. 6 | 7 | ## Bio 8 | Shashank Subramanian is a NESAP for learning postdoctoral fellow with research interests in the intersection of high-performance scientific computing, deep learning, and physical sciences. 9 | -------------------------------------------------------------------------------- /abstracts/2020-08-21.md: -------------------------------------------------------------------------------- 1 | # Weights & Biases: system of record to track, optimize, and reproduce ML research 2 | ## Chris Van Pelt and Charles Frye (Weights & Biases) 3 | 4 | ## Abstract 5 | Weights & Biases builds developer tools to track & monitor machine learning experiments, visualize metrics, and share results with rich dashboards and reports. The W&B team, including Chris Van Pelt (Co-Founder) and Charles Frye (Deep Learning Educator) will join us for an interactive product demo and discussion of the following topics: 6 | 7 | - How W&B ensures reproducibility by tracking and organizing experiments, datasets, and model versions--all in a unified system of record 8 | - W&B's Sweeps tool, which makes launching, monitoring, and recording large, distributed hyperparameter searches elegantly simple 9 | - W&B Reports, where researchers can create rich, interactive documents to showcase their work 10 | - W&B Product Roadmap 11 | - Next steps: getting started with W&B 12 | 13 | ## Bio 14 | Chris Van Pelt is a co-founder of Weights & Biases. In 2009, Chris founded Figure Eight/CrowdFlower. Over the past 10 years, Chris has dedicated his career to optimizing ML workflows and teaching ML practitioners, making machine learning more accessible to all. Chris has worked as a studio artist, computer scientist, and web engineer. He studied both art and computer science at Hope College. 15 | 16 | Charles Frye, PhD is a Deep Learning Educator at Weights & Biases. Charles did his thesis work on the loss surfaces of deep neural networks in the Redwood Center for Theoretical Neuroscience at the University of California, Berkeley with Professors Michael DeWeese and Kristofer Bouchard, graduating in 2020. 17 | -------------------------------------------------------------------------------- /abstracts/2020-06-05.md: -------------------------------------------------------------------------------- 1 | # Tuning Floating Point Precision (Using Dynamic and Temporal Locality Program Information) 2 | ## Costin Iancu (Computational Research Division, Lawrence Berkeley National Laboratory) 3 | 4 | ## Abstract 5 | Tuning the width of data used in scientific computations has been shown to lead to performance, energy and memory bandwidth improvements. For most of our codes, this is equivalent to tuning the precision of floating point operations. Given the current hardware specialization trends which often introduce fixed functional units on narrow data types, and the fact that extreme heterogeneity seems to be one of the ASCR strategic directions, we expect some application teams to try to tune the precision of floating point data used in their simulations. This talk summarizes the artifacts of a year long NERSC/CRD research collaboration. In the first part, we will discuss tools and techniques to reason about errors and their magnitude in scientific codes. In the second part of the talk we will introduce techniques we have developed to enable precision tuning of full application codes. Our automated procedure combines application “feature” extraction from execution traces with a hierarchical search strategy that combines static and dynamic (backtrace, temporal locality) information. When tuning two NERSC codes (CCTBX - Computational Crystallography Toolbox and AMReX-Combustion/PeleC) we attained performance improvements up to 40\% and were able to lower the precision of a good fraction of the total operation count. We will conclude with a discussion of several research directions opened by this preliminary investigation. 6 | 7 | ## Bios 8 | Costin Iancu is a Senior Staff Scientist in the Computational Research Division at Lawrence Berkeley National Laboratory 9 | -------------------------------------------------------------------------------- /abstracts/2022-03-22.md: -------------------------------------------------------------------------------- 1 | # Composable Platforms for Scientific Computing: Experiences and Outcomes 2 | ## Brian Werts, Sam Weekly and Erik Gough (Rosen Center for Advanced Computing, Purdue University) 3 | 4 | ## Abstract 5 | The Geddes Composable Platform is an on-premise Kubernetes-based private cloud hosted at Purdue University that’s designed to meet the increased demand for scientific data analysis and to promote "SciOps" — the application of DevOps principles in scientific computing. The platform has supported research groups and data science initiatives at Purdue, enabling as many as sixty users from a variety of scientific domains. In this seminar, we will give a technical overview of the platform and its components, summarize the usage patterns, and describe the scientific use cases the platform enables. Some examples of services deployed through Geddes include JupyterHubs, science gateways, databases, ML-based image classifiers, and web-based BLAST database searches. The same technology behind Geddes is found in Purdue’s new XSEDE resource named Anvil, which provides composable computing capabilities to the broader national research community. 6 | 7 | ## Bio 8 | Erik Gough is a lead computational scientist in the Research Computing department at Purdue University. He has been building, maintaining and using large scale cyberinfrastructure for scientific computing at Purdue since 2007. Gough is a technical leader on multiple NSF funded projects, including an NSF CC* award to build the Geddes Composable Platform. 9 | 10 | Brian Werts is the lead engineer for the design and implementation of Purdue’s Geddes Composable Platform and a HIPAA aligned Hadoop cluster for researchers that leverages Kubernetes to help facilitate reproducibility and scalability of data science workflows. 11 | -------------------------------------------------------------------------------- /abstracts/2019-05-21.md: -------------------------------------------------------------------------------- 1 | ## Cascade Reconstruction in IceCube using Convolutional and Generative Neural Networks 2 | ### Mirco Huennefeld (TU Dortmund) 3 | 4 | A key challenge to the success of high-energy physics experiments such as IceCube is the reliable and accurate reconstruction of events. In IceCube, further challenges arise as the detector is situated at the geographic South Pole where resources are limited. However, to perform real-time analyses and to issue alerts to telescopes around the world, powerful and fast reconstruction methods are desired. This results in a dilemma as performance is often paired with computational complexity. But even for offline reconstructions, the computational complexity of the most advanced maximum likelihood methods can render these intractable and hence limit the physics potential. Deep neural networks can be extremely powerful and their usage is computationally inexpensive once the networks are trained. These characteristics make a deep learning-based approach an excellent candidate for the application in IceCube. 5 | 6 | A reconstruction method based on convolutional architectures and hexagonally shaped kernels is presented. This method has been verified on real data and can significantly increase the reconstruction accuracy while reducing the runtime in comparison to standard reconstruction 7 | methods in IceCube. Although it can considerably improve the reconstruction performance, the presented CNN-based method has its limitations. In the typical physics use-case, many symmetries, invariances, and prior knowledge exist in the data, which are yet to be exploited by most standard network architectures. An approach using generative neural networks is introduced which has the potential to exploit this knowledge by combining strengths of neural networks and maximum likelihood methods. 8 | -------------------------------------------------------------------------------- /abstracts/2020-05-01.md: -------------------------------------------------------------------------------- 1 | # Deep learning for PDEs, and scientific computing with JAX 2 | ## Stephan Hoyer (Google) 3 | 4 | ## Abstract 5 | This talk will give an overview of how deep learning can be combined with traditional numerical methods to create improved methods for scientific computing. I will highlight two recent examples from my research: using deep learning to improve discretizations for solving partial differential equations [1], and using deep learning to reparameterize optimization landscapes for PDE constrained structural optimization [2]. I will also briefly introduce JAX [3], an open source library from Google for composable transformations of Python/NumPy programs, including automatic differentiation, vectorization and JIT compilation for accelerators. JAX is particularly suitable for scientific applications, including hybrid machine learning / simulation codes. 6 | 7 | [1] Bar-Sinai*, Y., Hoyer*, S., Hickey, J. & Brenner, M. P. Learning data-driven discretizations for partial differential equations. Proceedings of the National Academy of Sciences 201814058 (2019). doi:10.1073/pnas.1814058116 [2] Hoyer, S., Sohl-Dickstein, J. & Greydanus, S. Neural reparameterization improves structural optimization. arXiv [cs.LG] (2019). https://arxiv.org/abs/1909.04240 [3] https://github.com/google/jax 8 | 9 | 10 | ## Bio 11 | Stephan is a researcher and software engineer at Google on the Accelerated Sciences team (https://research.google/teams/applied-science/gas/), where he works on applications of machine learning to scientific computing and the physical sciences. Before Google, he worked on machine learning for weather forecasting at The Climate Corporation and completed a Physics PhD at UC Berkeley. He is also a frequent contributor to open source projects in the Python scientific computing stack, including NumPy, Dask, Xarray, and JAX. 12 | -------------------------------------------------------------------------------- /abstracts/2021-10-19.md: -------------------------------------------------------------------------------- 1 | # Scaling Out HPC with On-Premise Performance in the Oracle Cloud Infrastructure 2 | ## Luiz DeRose, Ph.D. (Oracle) 3 | 4 | ## Abstract 5 | The continuous increase in complexity and scale of high-end systems, together with the evolving diversity of processor options, are forcing computational scientists to face system characteristics that can significantly impact the performance and scalability of applications. HPC users need a system infrastructure that can adapt to their workload needs, rather than having to constantly redesign their applications to adapt to new systems. In this talk, I will discuss the current trends in computer architecture and the implications in the development of HPC applications and programming and middleware environments. I will present the Oracle Cloud Infrastructure (OCI), which provides availability, resiliency, and performance at scale, so HPC users can easily choose the best option for their workloads, and will discuss hybrid on-prem/cloud options, which facilitate workload migration from on-premise to the cloud. I will finish the presentation with a discussion of some of the challenges and open research problems that still need to be addressed in this area. 6 | 7 | ## Bio 8 | Dr. Luiz DeRose is a Director of Cloud Engineering for HPC at Oracle. Before joining Oracle, he was a Sr. Science Manager at AWS, and a Senior Principal Engineer and the Programming Environments Director at Cray. Dr. DeRose has a Ph.D. in Computer Science from the University of Illinois at Urbana-Champaign. He has more than 25 years of high-performance computing experience and a deep knowledge of programming and middleware environments for HPC. Dr. DeRose has eight patents and has published more than 50 peer-review articles in scientific journals, conferences, and book chapters, primarily on the topics of compilers and tools for high performance computing. 9 | -------------------------------------------------------------------------------- /abstracts/2021-12-06.md: -------------------------------------------------------------------------------- 1 | # funcX: Federated FaaS for Scientific Computing 2 | ## Ryan Chard (Data Science and Learning Division - Argonne National Laboratory) 3 | 4 | ## Abstract 5 | Exploding data volumes and velocities, new computational methods and platforms, and ubiquitous connectivity demand new approaches to computation in the sciences. These new approaches must enable computation to be mobile, so that, for example, it can occur near data, be triggered by events (e.g., arrival of new data), be offloaded to specialized accelerators, or run remotely where resources are available. They also require new design approaches in which monolithic applications can be decomposed into smaller components, that may in turn be executed separately and on the most suitable resources. To address these needs we present funcX—a distributed function as a service (FaaS) platform that enables flexible, scalable, and high-performance remote function execution. funcX's endpoint software can transform existing clusters and supercomputers into function serving systems, while funcX's cloud-hosted service provides transparent, secure, and reliable function execution across a federated ecosystem of endpoints. We demonstrate the use of funcX with several scientific case studies and show how it integrates into the wider Globus ecosystem to enable secure, fire-and-forget scientific computation. 6 | 7 | ## Bio 8 | Ryan Chard joined Argonne in 2016 as a Maria Goeppert Mayer Fellow and then as an Assistant Computer Scientist in the Data Science and Learning Division. He now works with Argonne, UChicago, and Globus to develop cyberinfrastructure to enable scientific research. In particular, he works on the Globus Flows platform to create reliable data analysis pipelines and the funcX service to enable function serving for HPC. He has a Ph.D. in computer science and an M.Sc. from Victoria University of Wellington, New Zealand. 9 | -------------------------------------------------------------------------------- /abstracts/2022-04-19.md: -------------------------------------------------------------------------------- 1 | # Discovering and Modeling Strong Gravitational Lenses with Cori and Perlmutter at NERSC 2 | ## Xiaosheng Huang (USF), Andi Gu (UCB) 3 | 4 | ## Abstract 5 | We have discovered over 1500 new strong lens candidates in the Dark Energy Spectroscopic Instrument (DESI) Legacy Imaging Surveys with residual neural networks using NERSC resources. Follow-up observations are underway. Our Hubble Space Telescope program has confirmed all 51 observed candides. DESI observations have confirmed more systems spectroscopically. Preliminary results from our latest search will increase the number of lens candidates to over 3000. We have also developed GIGA-Lens: a gradient-informed, GPU-accelerated Bayesian framework, implemented in TensorFlow and JAX. All components of this framework (optimization, variational inference, HMC) take advantage of gradient information through autodiff and parallelization on GPUs. Running on one Perlmutter A100 GPU node, we achieve 1-2 orders of magnitude speedup compared to existing codes. The robustness, speed, and scalability offered by this framework make it possible to model the large number of strong lenses found in DESI, and O(10^5) lenses expected to be discovered in upcoming large-scale surveys, such as the LSST. 6 | 7 | ## Bios 8 | Xiaosheng Huang received his PhD from UC Berkeley and has been a faculty member in the Physics & Astronomy Department at the University of San Francisco since 2012. He works on problems in observational cosmology with collaborators in the Supernova Cosmology Project, the Nearby Supernova Factory, and the Dark Energy Spectroscopic Instrument experiment, and of course, with students. 9 | 10 | Andi Gu is a current senior at UC Berkeley. He has been working in the Supernova Cosmology Project and DESI since 2019, applying his computer science and physics background to gravitational lens detection and modeling. 11 | -------------------------------------------------------------------------------- /abstracts/2020-07-17.md: -------------------------------------------------------------------------------- 1 | # Superconducting Radio-Frequency Cavity Fault Classification Using Machine Learning at Jefferson Laboratory 2 | ## Christopher Tennant (Jefferson Laboratory) 3 | 4 | ## Abstract 5 | We report on the development of machine learning models for classifying C100 superconducting radio-frequency (SRF) cavity faults in the Continuous Electron Beam Accelerator Facility (CEBAF) at Jefferson Lab. CEBAF is a continuous-wave recirculating linac utilizing 418 SRF cavities to accelerate electrons up to 12 GeV through 5-passes. Of these, 96 cavities (12 cryomodules) are designed with a digital low-level RF system configured such that a cavity fault triggers waveform recordings of 17 RF signals for each of the 8 cavities in the cryomodule. Subject matter experts (SME) are able to analyze the collected time-series data and identify which of the eight cavities faulted first and classify the type of fault. This information is used to find trends and strategically deploy mitigations to problematic cryomodules. However manually labeling the data is laborious and time-consuming. By leveraging machine learning, near real-time – rather than post-mortem – identification of the offending cavity and classification of the fault type has been implemented. We discuss the development and performance of the ML models as well as valuable lessons learned in bringing a ML system to deployment. 6 | 7 | ## Bio 8 | Chris Tennant is an accelerator physicist who earned his Ph.D. while doing research at Jefferson Lab in 2006, and has been employed as staff scientist ever since. For most of his career his research has been in the area of energy recovery linacs for a broad range of applications; from electron cooling for electron-ion colliders, to free-electron lasers for lithography and defense. More recently his interests have shifted toward practical applications of machine learning to improve beam operations. 9 | -------------------------------------------------------------------------------- /abstracts/2021-01-19.md: -------------------------------------------------------------------------------- 1 | # Self-Supervised Representation Learning for Astronomical Images 2 | ## Md Abul Hayat (University of Arkansas, Berkeley Lab), George Stein (UCB, Berkeley Lab) 3 | 4 | ## Abstract 5 | Sky surveys are the largest data generators in astronomy, making automated tools for extracting meaningful scientific information an absolute necessity. We show that, without the need for labels, self-supervised learning recovers representations of sky survey images that are semantically useful for a variety of scientific tasks. These representations can be directly used as features, or fine-tuned, to outperform supervised methods trained only on labeled data. We apply a contrastive learning framework on multi-band galaxy photometry from the Sloan Digital Sky Survey (SDSS), to learn image representations. We then use them for galaxy morphology classification, and fine-tune them for photometric redshift estimation, using labels from the Galaxy Zoo 2 dataset and SDSS spectroscopy. In both downstream tasks, using the same learned representations, we outperform the supervised state- of-the-art results, and we show that our approach can achieve the accuracy of supervised models while using 2-4 times fewer labels for training. 6 | 7 | ## Bio 8 | Md Abul Hayat is a 4th year PhD student of Electrical engineering at the University of Arkansas. His research is on predicting hydration status using signal processing and statistical learning techniques from biomedical signals. In past years he has worked at Berkeley Lab and Nokia Bell Labs as a summer intern. Before joining the PhD program, he has worked as a telecommunications system engineer at Telenor Bangladesh. 9 | 10 | George Stein is a postdoc at LBL/BCCP, with research centered on machine learning for cosmology. Areas of focus include cosmological simulations, generative models, anomaly detection, and of course self-supervised learning. 11 | -------------------------------------------------------------------------------- /abstracts/2022-01-13.md: -------------------------------------------------------------------------------- 1 | # The Next Generation of High Performance Computing: HPC-2.0 2 | ## Gregory Kurtzer, CEO of Ctrl IQ, Inc. and Executive Director of Rocky Enterprise Software Foundation / Rocky Linux 3 | 4 | ## Abstract 5 | We’ve been using the same base architecture for building HPC systems for almost 30 years and while the capabilities of our systems have increased considerably, we still use the same flat and monolithic architecture of the 1990’s to build our systems. What would the next generation architecture look like? How do we leverage containers to do computing of complex workflows while orchestrating not only jobs, but data? How do we bridge HPC into the 2020’s and make optimal use of multi-clusters and federate these systems into a larger resource to unite on-prem, multi-prem, cloud, and multi-cloud? How do we integrate with these resources in a cloud-native compatible manner supporting CI/CD, DevOps, DevSecOps, compute portals, GUIs, and even mobile? This isn’t a bunch of shoelace and duct-tape on top of legacy HPC, this is an entirely new way to think about HPC infrastructure. This is a glimpse into HPC-2.0, coming later in Q1 of 2022. 6 | 7 | ## Bio 8 | Gregory M. Kurtzer is a 20+ year veteran in Linux, open source, and high performance computing. He is well known in the HPC space for designing scalable and easy to manage secure architectures for innovative performance intensive computing while working for the U.S. Department of Energy and joint appointment to UC Berkeley. Greg founded and led several large open source projects such as CentOS Linux, the Warewulf and Perceus cluster toolkits, the container system Singularity, and most recently, the successor to CentOS, Rocky Linux. Greg’s first startup was acquired almost 2 years ago and now he is working on software infrastructure, including Rocky Linux as well as building a cloud native, cloud hybrid, federated orchestration platform called Fuzzball. 9 | -------------------------------------------------------------------------------- /abstracts/2021-01-12.md: -------------------------------------------------------------------------------- 1 | # Deep Learning Approaches for Modeling Multi-Scale Chaos and Geophysical Turbulence 2 | ## Ashesh Chattopadhyay (Rice University) 3 | 4 | ## Abstract 5 | Our atmosphere is a coupled, chaotic, and turbulent dynamical system with 6 | multiple physical processes interacting with each other, at continuously varying spatio-temporal scales. Building efficient and accurate weather/climate models that can predict the state of the atmosphere for the near and distant future requires us to resolve a broad range of spatio-temporal scales that often take up a daunt- 7 | ing amount of computational resources. Thus, current tractable climate models often have inaccurate and crude approximations of hard-to-resolve physical processes that drastically affect our ability to predict the dynamics of the system. Here, we propose alternative data-driven approaches that utilize deep learning algorithms trained on observations or high-resolution model outputs working in conjunction with numerical models to perform carefully constructed approximations that accurately capture the physics of these hard-to-resolve processes. This can reduce computational cost while bringing more insight into poorly understood physics that can dramatically improve our ability to predict the large-scale dynamics of the atmosphere. 8 | 9 | ## Bio 10 | Ashesh Chattopadhyay did his Bachelors from the department of Mechanical Engineering at Indian Institute of Technology, Patna where he worked primarily in optimization and computational geometry. He got his masters from the University of Texas, El Paso, from the Computational Science program where his research was focused on high performance computing. Since then, he has been a PhD student at Rice University in the department of Mechanical Engineering, where he works at the intersection of theoretical deep learning, dynamical systems and turbulence modeling for broad applications in atmospheric dynamics. 11 | -------------------------------------------------------------------------------- /abstracts/2020-10-16.md: -------------------------------------------------------------------------------- 1 | # Using Machine Learning to Augment Coarse-Grid Computational Fluid Dynamics Simulations 2 | ## Jaideep Pathak (NERSC) 3 | 4 | ## Abstract 5 | Simulation of turbulent flows at high Reynolds number is a computationally challenging task relevant to a large number of engineering and scientific applications in diverse fields such as climate science, aerodynamics, and combustion. Turbulent flows are typically modeled by the Navier-Stokes equations. Direct Numerical Simulation (DNS) of the Navier-Stokes equations with sufficient numerical resolution to capture all the relevant scales of the turbulent motions can be prohibitively expensive. Simulation at lower-resolution on a coarse-grid introduces significant errors. We introduce a machine learning (ML) technique based on a deep neural network architecture that corrects the numerical errors induced by a coarse-grid simulation of turbulent flows at high-Reynolds numbers, while simultaneously recovering an estimate of the high-resolution fields. Our proposed simulation strategy is a hybrid ML-PDE solver that is capable of obtaining a meaningful high-resolution solution trajectory while solving the system PDE at a lower resolution. The approach has the potential to dramatically reduce the expense of turbulent flow simulations. As a proof-of-concept, we demonstrate our ML-PDE strategy on a two-dimensional Rayleigh-Bénard Convection (RBC) problem. 6 | 7 | 8 | ## Bio 9 | Jaideep Pathak is a NESAP for Learning Postdoctoral Fellow at NERSC, Lawrence Berkeley National Laboratory working on incorporating machine learning learning techniques for problems in computational fluid dynamics. He joined NERSC after receiving his PhD from the University of Maryland, College Park in December 2019. His primary interests are in developing machine learning techniques for improving simulations in diverse fields such as weather forecasting and climate modeling, fluid turbulence, and combustion. 10 | -------------------------------------------------------------------------------- /abstracts/2020-05-29.md: -------------------------------------------------------------------------------- 1 | # Simulation-based and label-free deep learning for science 2 | ## Ben Nachmann (Lawrence Berkeley National Laboratory) 3 | 4 | ## Abstract 5 | Precise scientific analysis in many areas of science is possible because of complex simulations that connect fundamental theories to observable quantities. These simulations have been paired with multivariate methods for many years in search of new fundamental and emergent structure in nature. Deep learning tools hold great promise to qualitatively change this paradigm by allowing for holistic analysis of data in its natural hyperdimensionality with thousands or millions of features instead of up to tens of features. These tools are not yet broadly used for all areas of data analysis because of the traditional dependence on simulations. In this talk, I will discuss how we can change this paradigm in order to exploit the new features of deep learning. In particular, I will show how neural networks can be used to (1) overcome the challenge of high-dimensional probability density modeling and (2) learn directly from (unlabeled) data to perform hypothesis tests that go beyond any existing analysis methods. The example for (1) will be full phase space unfolding (deconvolution) and the example for (2) will be anomaly detection. The talk will include a discussion of uncertainties associated with deep learning-based analyses. These ideas are starting to become a reality: the first deep learning weakly supervised anomaly detection search has recently been made public by the ATLAS Collaboration at the LHC. While my examples will primarily draw from collider physics, the techniques are more broadly applicable and I am happy to discuss extensions and applications to your science domain. 6 | 7 | ## Bios 8 | Benjamin Nachman is currently Chamberlain Fellow at the Lawrence Berkeley National Laboratory and a member of the ATLAS Collaboration at CERN. He obtained his PhD in Physics, with a minor in Statistics, from Stanford University. 9 | -------------------------------------------------------------------------------- /abstracts/2021-03-09.md: -------------------------------------------------------------------------------- 1 | # Darshan: Enabling Application I/O Understanding in an Evolving HPC Landscape 2 | ## Shane Snyder (Argonne National Laboratory) 3 | 4 | ## Abstract 5 | Darshan is a lightweight, application I/O characterization tool that captures detailed statistics describing an application's I/O workload. Installed and enabled by default at many production HPC facilities (including at NERSC), Darshan has become an invaluable tool for users, system admins, and I/O researchers to investigate and tune the I/O behavior of applications. While the initial focus of Darshan was on instrumenting file-based APIs (e.g., POSIX, MPI-IO) for MPI applications, much recent work has focused on extending Darshan to new contexts that are increasingly relevant in the HPC community, including object-based storage APIs (e.g., DAOS) and non-MPI computational frameworks (e.g., Spark, TensorFlow). In this seminar, we describe how users can leverage Darshan to better understand the I/O behavior of their applications. We provide details on how users can produce Darshan instrumentation data for their applications and how to further analyze this data, focusing specifically on the Cori system at NERSC. New and upcoming features are covered that aim to extend Darshan to exciting I/O instrumentation contexts for HPC, including instrumentation modules for HDF5 and DAOS libraries, as well as support for instrumenting non-MPI applications and frameworks. We further walk through a couple of Darshan log analysis examples to help illustrate the types of I/O insights that can be attained using Darshan log data and analysis tools. 6 | 7 | ## Bio 8 | Shane Snyder is a software engineer in the Mathematics and Computer Science Division of Argonne National Laboratory. He received his master's degree in computer engineering from Clemson University in 2013. His research interests primarily include the design of high-performance distributed storage systems and the characterization and analysis of I/O workloads on production HPC systems. 9 | -------------------------------------------------------------------------------- /abstracts/2022-06-14.md: -------------------------------------------------------------------------------- 1 | # Artificial Design of Porous Materials 2 | ## Jihan Kim, Department of Chemical and Biomolecular Engineering, KAIST 3 | 4 | ## Abstract 5 | In this presentation, I will explore the new trend of designing novel porous materials using artificial design principles. I will talk about using our in-house developed generative adversarial network (GAN) software to create (for the first time) porous materials. Moreover, we have successfully implemented inverse-design in our GAN prompting ways to train our AI to create porous materials with user-desired methane adsorption capacity [1]. Next, we incorporate machine learning with genetic algorithm to design optimal metal-organic frameworks suitable for many different applications including methane storage and gas separations [2-3]. Finally, we demonstrate usage of text mining to collect wealth of data from published papers to predict optimal synthesis conditions for porous materials [4]. Overall, machine learning and artificial design can accelerate the materials discovery and expedite the process to deploy new materials for many different applications. 6 | 7 | ## Bio 8 | Jihan Kim is an associate professor at KAIST (Korea Advanced Institute of Science and Technology). He received his B.S. degree in Electrical Engineering and Computer Science (EECS) at UC Berkeley in 1997 and received his M.S. and Ph.D. degrees in Electrical and Computer Engineering at University of Illinois at Urbana-Champaign in 2004 and 2009, respectively. He worked as a NERSC postdoc in the Petascale Post-doc project from 2009 to 2011 and worked as postdoctoral researcher in UC Berkeley/LBNL with Prof. Berend Smit from 2011 to 2013. His current research at KAIST focuses on using molecular simulations and machine learning methods to design novel porous materials (e.g. zeolites, MOFs, porous polymers) for various energy and environmental related applications (e.g. gas storage, gas separations, catalysis, sensors). He has published over 100 papers and has over 7000 Google Scholar citations. 9 | -------------------------------------------------------------------------------- /abstracts/2020-06-26.md: -------------------------------------------------------------------------------- 1 | # Congestion and Distributed Training of Deep Neural Networks 2 | ## Jacob Balma (HPE) 3 | 4 | ## Abstract 5 | Congestion is one of the biggest problems facing large-scale multi-user HPC systems today. It results in fewer jobs being run per dollar spent over the life of the system, affecting not only the system’s total throughput, but also user experience. Loaded systems experience congestion in the form of run-to-run variability and contention for shared resources like filesystems or routes between compute endpoints. Emerging workloads in data-science and specifically deep learning can exacerbate contention for shared resources by requiring high-bandwidth between I/O subsystems and compute nodes simultaneously. Current network benchmarks are not capable of creating the conditions necessary to proxy real-world network utilization seen on congested systems, nor do they offer mechanisms for quantifying the impact it has on user applications which run under such conditions. A recently proposed open-source set of network benchmarks called the Global Performance and Congestion Network Tests (GPCNeT) are aimed at providing an industry standard for performance characterization of well-utilized HPC networks. We expand the GPCNeT benchmark suite to support our investigation of the communication patterns intrinsic to Synchronous Distributed SGD, commonly used to train Deep Neural Networks (DNNs) on distributed systems. 6 | 7 | We show that Synchronous Distributed SGD can cause Congestion when run alongside the Congestion-Sensitive algorithms used by GPCNeT to proxy prominent features associated with many common HPC workloads. Finally, we compare the degree to which hyper-parameter tuning algorithms when applied to Synchronous Distributed SGD affect the congestion impact factor relative to various sensitive workloads. 8 | 9 | ## Bio 10 | Jacob Balma works as an HPC & AI Research Scientist at HPE. His background is in statistical physics, scientific simulation and benchmarking. Current interests include drug discovery, open-therapeutics and materials science. 11 | -------------------------------------------------------------------------------- /abstracts/2020-11-02.md: -------------------------------------------------------------------------------- 1 | # Robust prediction of high-dimensional dynamical systems using Koopman deep networks 2 | ## Omri Azencot (Ben-Gurion University) 3 | 4 | ## Abstract 5 | We present a new deep learning approach for the analysis and processing of time series data. At the core of our work is the Koopman operator which fully encodes a nonlinear dynamical system. Unlike the majority of Koopman-based models, we consider dynamics for which the Koopman operator is invertible. We exploit the structure of these systems to design a novel Physically-Constrained Learning (PCL) model that takes into account the inverse dynamics while penalizing for inverse prediction. Our architecture is composed of an autoencoder component and two Koopman layers for the dynamics and their inverse. To motivate our network design, we investigate the connection between invertible Koopman operators and pointwise maps, and our analysis yields a loss term which we employ in practice. To evaluate our work, we consider several challenging nonlinear systems including the pendulum, fluid flows on curved domains and real climate data. We compare our approach to several baseline methods, and we demonstrate that it yields the best results for long time predictions and in noisy settings. 6 | 7 | ## Bio 8 | Omri Azencot is a Senior Lecturer (Assistant Professor) in the Computer Science Department at the Ben-Gurion University of the Negev, Israel. His research interests include deep learning, sequence models, and dynamical systems. Omri did his postdoc at the Department of Mathematics at the University of California, Los Angeles, hosted by Prof. Andrea Bertozzi. He completed his PhD at the Computer Science Department at the Technion -- Israel Institute of Technology, under the supervision of Prof. Mirela Ben-Chen. Prior to that, he obtained a BSc in computer science and a BSc in mathematics, both from the Technion. Omri’s research has been supported by an Adams Fellowship of the Israel Academy of Sciences and Humanities, a Zuckerman Postdoctoral Fellowship, and a Marie Sklodowska-Curie Actions International Fellowship. 9 | -------------------------------------------------------------------------------- /abstracts/2021-12-07.md: -------------------------------------------------------------------------------- 1 | # Ceph Storage at CERN 2 | ## Dan van der Ster, Pablo Llopis Sanmillan (CERN) 3 | 4 | ## Abstract 5 | Ceph and its Reliable Autonomic Distributed Object Store (RADOS) offers a scale out storage solution for block storage (RBD), object storage (S3 and SWIFT) and filesystems (CephFS). The key technologies enabling Ceph include CRUSH, a mechanism for defining and implementing failure domains, and the mature Object Storage Daemons (OSDs), which provide a reliable storage backend via replication or erasure coding. CERN has employed Ceph for its on-prem cloud infrastructures since 2013. As of 2021, its storage group operates more than ten clusters totaling over 50 petabytes for cloud, Kubernetes, and HPC use-cases. This talk will introduce Ceph and its key concepts, and describe how CERN uses Ceph in practice. It will include recent highlights related to high-throughput particle physics data taking and SLURM storage optimization. 6 | 7 | ## Bio(s) 8 | Dan manages the Ceph storage at CERN in Geneva, Switzerland. He has participated actively in its community since 2013 and was one of the first to demonstrate its scalability up to multi-10s of petabytes. Dan is a regular speaker at Open Infrastructure events, previously acted as Academic Liaison to the original Ceph Advisory Board, and has a similar role in the current Ceph Board. Dan earned a PhD in Distributed Systems at the University of Victoria, Canada in 2008. 9 | 10 | Pablo is a computer engineer at CERN, where he manages the IT department’s HPC service. He provides HPC support to both engineers of the Accelerator Technology Sector and to theoretical physicists. Pablo works on improving the performance of their HPC workloads, and on other projects such as the automation of operational tasks of the infrastructure. He holds a Ph.D. in computer science from University Carlos III of Madrid. In the past he has also collaborated with Argonne National Laboratory and IBM Research Zurich on HPC and cloud-related topics. His main areas of interest include high performance computing, storage systems, power efficiency, and distributed systems. 11 | -------------------------------------------------------------------------------- /abstracts/2020-05-15.md: -------------------------------------------------------------------------------- 1 | # Deep Learning Production Capabilities at NERSC 2 | ## Steven Farrell, Mustafa Mustafa (NERSC) 3 | 4 | ## Abstract 5 | Deep Learning is increasingly being used for scientific problems which require large scale computing resources. High performance computing centers are adapting to accommodate these new kinds of workloads which can differ significantly from traditional HPC simulation workloads. NERSC supports and enables deep learning workflows by providing an optimized software stack and by supporting users to deploy their applications effectively and productively. In this presentation we will describe NERSC’s production capabilities for scientific deep learning applications, including details of the software stack, system performance with extensive benchmarking, and workflow solutions to enable productive science. In addition, we will discuss our outlook for the future of AI at NERSC on the upcoming Perlmutter supercomputer and beyond. 6 | 7 | 8 | ## Bios 9 | Steven Farrell is a Machine Learning Engineer at NERSC. He supports scientific deep learning workflows on HPC systems through software development, benchmarking, user support, and training. His research interests include applications of deep learning to high energy physics, generative modeling, and applications of learning on structured data such as graphs. Steve is co-chair of the MLPerf HPC working group. He was a member of the ATLAS experiment at CERN for many years, first during his Ph.D studies at UC Irvine and then as a postdoc at Berkeley Lab working on software development and machine learning applications for analysis and simulation. 10 | 11 | Mustafa Mustafa is a Machine Learning Engineer at NERSC supercomputing center. His current interests are in deep learning optimization at scale and data-driven physics modeling (spatio-temporal generative models for surrogate modeling, generative model training dynamics and application to scientific problems). He enjoys communicating science and lecturing on deep learning and how it works. Mustafa's background is in experimental high energy nuclear physics, he obtained his Ph.D. from Purdue University and was a Postdoctoral Fellow at the Nuclear Science Division at Berkeley Lab. 12 | -------------------------------------------------------------------------------- /abstracts/2021-11-02.md: -------------------------------------------------------------------------------- 1 | # Characterizing I/O behavior of large-scale scientific deep learning applications 2 | ## Hariharan Devarajan (Lawrence Livermore National Laboratory) 3 | 4 | ## Abstract 5 | Deep learning has been shown as a successful method for various tasks, and its popularity results in numerous open-source deep learning software tools. Deep learning has been applied to a broad spectrum of scientific domains such as cosmology, particle physics, computer vision, fusion, and astrophysics. Scientists have performed a great deal of work to optimize the computational performance of deep learning frameworks. However, the same cannot be said for I/O performance. As deep learning algorithms rely on big-data volume and variety to effectively train neural networks accurately, I/O is a significant bottleneck on large-scale distributed deep learning training. In this talk, I will share our experiences of running large-scale DL applications on Theta supercomputer with a detailed investigation of the I/O behavior of various scientific deep learning workloads. Additionally, I will showcase our DLIO Benchmark, which accurately represents the class of applications previously characterized to foster I/O research in these classes of applications. I will share some key results and insights we discovered in modern scientific DL applications including, access patterns, integration with scientific data formats, and their I/O scalability in production supercomputers. Finally, I would highlight key pain points in doing I/O characterization of DL applications and discuss some research directions to improve these aspects. 6 | 7 | ## Bio 8 | Hariharan Devarajan is a Postdoctoral researcher at Lawrence Livermore National Laboratory. He received his Ph.D. in Computer Science at Illinois Institute of Technology, advised by Dr. Xian-He Sun. His research is focused on accurate I/O characterization of distributed applications and building highly configurable storage systems on large-scale distributed systems. He has worked on I/O optimizations in several domains such as scientific simulations, AI, and Big Data Analytics and specializes in designing solutions for hierarchical storage environments. He is the recipient of the best paper awards at HPDC and CCGrid. 9 | -------------------------------------------------------------------------------- /abstracts/2020-01-31.md: -------------------------------------------------------------------------------- 1 | # Data skeletons: IO workload characterization for the modern age 2 | ## Avani Wildani (Emory University) 3 | 4 | ## Abstract 5 | In a not-too-distant future, our devices will not just be networked, but share huge volumes of data required for instantaneous and vital decisions. Technologies such as self-driving cars, smart cities and homes, and augmented reality all depend on handling massive quantities of data quickly and reliably. Understanding the characteristics of workloads such as these will be critical to reducing storage costs and making the cloud of tomorrow accessible and equitable for all. All storage systems exist to serve some group of users and applications. Tuning a storage system is a delicate balance between reliability, availability, security, performance spread over the users, or workloads, that the system serves. For cloud storage, elasticity is a crucial factor, but misconfiguration in reactive storage tuning has been cited to be a leading cause of production failures. Transferring and transforming provisioning insights to match dynamic workloads will break ground for system improvements spanning from the power footprint to cache management to selecting an appropriate reliability configuration. This talk covers the current state of workload-aware design along with our current work to improve storage provisioning and trace characterization. 6 | 7 | ## Bio: 8 | Dr. Avani Wildani is an Assistant Professor in Computer Science and Neuroscience and co-PI of the SimBioSys lab at Emory University. Her group focuses on information models in cloud and communication systems, particularly those with biological connections, with a long term goal of categorizing neural information. Prior to that, she was a Pioneer Postdoctoral Fellow in computational neuroscience at the Salk Institute for Biological Sciences. She earned her B.S. in Computer Science and Mathematics at Harvey Mudd College and her Ph.D. in Computer Science at UC Santa Cruz under Dr. Ethan Miller. Her interests are centered around information storage and retrieval across different storage models, with application domains including access prediction, data deduplication, archival economics, power management, wireless mesh networks, auditory receptive field characterization, and pollution monitoring. 9 | -------------------------------------------------------------------------------- /abstracts/2021-08-10.md: -------------------------------------------------------------------------------- 1 | # LDMS data at NERSC: doing something useful with 8 petabytes of CSV files 2 | ## Brandon Cook (NERSC/LBL) 3 | 4 | ## Abstract 5 | Analysis of telemetry data from NERSC systems offers the potential for deeper quantitative understanding of the NERSC workload - providing insights for future system design, operation and optimization of the current platform, feedback to developers about workflow performance and diagnostics to uncover issues with workflows. NERSC uses the Lightweight Distributed Metric Service (LDMS) for lightweight collection of a large variety of metrics on NERSC systems; including memory usage, CPU HW counters, power consumption, network and I/O traffic. Across the compute nodes of Cori currently a total of ~400 MB/s worth of data is being collected and stored in CSV file format. With current retention policies there are approximately 8 petabytes of data in CSV format. The size and number of the CSV files along with the desire to integrate with other sources such as Slurm accounting poses several challenges for anyone who wants to work with this data. In this talk, I will walk through this data set: how it is collected, what is in it, where it is located. Then I will discuss a post processing pipeline that transforms, filters, joins this data with Slurm accounting information, and finally stores it 20 - 200 times more efficiently than CSV. Finally I will discuss how the results of the pipeline are used in Iris through the Superfacility API to provide plots directly to users for all non-shared jobs on Cori in O(seconds). Throughout the talk I will highlight how these resources can be accessed and extended. 6 | 7 | ## Bio 8 | Brandon leads the simulations area of NERSC's application readiness program (NESAP) and works on understanding and analyzing performance on a system and application level, developing future benchmark suites, analyzing future architectures, developing tools to help NERSC users/staff be more productive, engaging users through consulting, acting as NERSC liaison for several NESAP teams, and exploring future programming models. Brandon received his Ph.D. in physics from Vanderbilt University in 2012, where he studied ab initio methods for quantum transport in nanomaterials. Before joining NERSC he was a postdoc at Oak Ridge National Laboratory where he developed and applied electronic structure methods to problems in material science. 9 | -------------------------------------------------------------------------------- /abstracts/2020-09-11.md: -------------------------------------------------------------------------------- 1 | # Flux: Overcoming Scheduling Challenges for Exascale Workflows 2 | ## Dong Ahn & Stephen Herbein (Lawrence Livermore National Laboratory) 3 | 4 | ## Abstract 5 | Many emerging scientific workflows that target high-end HPC systems require complex interplay with the resource and job management software (RJMS). However, portable, efficient and easy-to-use scheduling and execution of these workflows is still an unsolved problem. In this talk, I will present Flux, a next-generation RJMS designed specifically to address the key scheduling challenges of modern workflows in a scalable, easy-to-use, and portable manner. At the heart of Flux lies its ability to be seamlessly nested within batch allocations created by itself as well as other system schedulers (e.g., SLURM, MOAB, LSF, etc), serving the target workflows as their “personal RJMS instances”. In particular, Flux’s consistent and rich set of well-defined APIs portably and efficiently support those workflows that can often feature non-traditional execution patterns such as requirements for complex co-scheduling, massive ensembles of small jobs and coordination among jobs in an ensemble. As part of this talk, I will also discuss Flux’s graph-based resource data model, Flux’s response to needing to schedule increasingly diverse resources, and how this model is becoming the center of our industry co-design efforts: for example, multi-tiered storage scheduling co-design with HPE and Cloud resource co-design with IBM T.J. Watson and RedHat OpenShift. 6 | 7 | 8 | ## Bio 9 | Dong H. Ahn is a computer scientist. He has worked for Livermore Computing (LC) at Lawrence Livermore National Laboratory since 2001 and currently leads the next-generation computing enabling (NGCE) project within the ASC ATDM sub-program. During this period, Dong has worked on several code development-tools and resource management and scheduling software framework projects with a common goal to provide highly capable and scalable tools ecosystems for large computing systems. Towards this goal, he has architected an extreme-scale debugging strategy that conceived the Stack Trace Analysis Tool (STAT), a 2011 R&D 100 Award winner, and the PRUNERS Toolset, a 2017 R&D 100 Award Finalist. 10 | 11 | Stephen Herbein is a computer scientist in Livermore Computing at Lawrence Livermore National Laboratory. His research interests include batch job scheduling, parallel IO, and data analytics. He is a part of the Flux team, developing next-generation IO-aware and multi-level schedulers for HPC. 12 | -------------------------------------------------------------------------------- /abstracts/2020-03-06.md: -------------------------------------------------------------------------------- 1 | # Intersections of AI/ML and Chemistry in Catalyst Design and Discovery 2 | ## Zachary Ulissi (UC Davis) 3 | 4 | ## Abstract 5 | 6 | Summary: Increasing computational sophistication and resources can enable a larger and more integrated role of theory in the discovery and understanding of new materials. This process has been slower to infiltrate surface science and catalysis than the field of bulk inorganic materials due to additional scientific complexity of modeling the interface. Most catalyst studies start in a data-poor regime where the material of interest is unrelated to previous to studies (new structure, composition etc) or the computational methods are incompatible with previous studies (different exchange-correlation functionals, methods, etc). Efficient methods to quickly define, schedule, and organize necessary simulations are thus important and enable the application of online design of experiments approaches. I will discuss on-going work and software development to enable data science methods in catalysis including open datasets for the community. These large datasets enable the use of graph convolutional models for surface properties and the uncertainty in these methods can be carefully calibrated. Finally, I will describe applications of our approach to ordered bimetallic alloy catalysts, with applications to several electrochemical catalyst discovery efforts including CO2 reduction, oxygen reduction, and water splitting chemistry. 7 | 8 | 9 | ## Bio 10 | Zachary Ulissi is an Assistant Professor in Chemical Engineering at Carnegie Mellon University in Pittsburgh PA. He did his undergraduate work in Chemical Engineering and Physics at the University of Delaware, a Masters in Applied Mathematics at Churchill College, Cambridge, and his PhD in Chemical Engineering at MIT funded by the DOE CSGF fellowship. His PhD research at MIT focused on the the application of systems engineering methods to understand selective nanoscale carbon nanotube devices and sensors working with Michael Strano and Richard Braatz. Prof. Ulissi did his postdoctoral work at Stanford with Jens Nørskov where he worked on machine learning techniques to simplify complex catalyst reaction networks and discover new catalysts. Current research interests include the design of high-throughput computational systems for surface science and catalysis, as well as the development of machine learning models and design of experiments processes to accelerate the discovery process. Methods include both all-atom classical (MD) and electronic (DFT) simulations techniques. 11 | -------------------------------------------------------------------------------- /abstracts/2021-12-14.md: -------------------------------------------------------------------------------- 1 | # Characterizing Machine Learning I/O Workloads on Leadership Scale HPC Systems 2 | ## Ahmad Maroof Karimi (NCCS - ORNL) 3 | 4 | ## Abstract 5 | High performance computing (HPC) is no longer solely limited to traditional workloads such as simulation and modeling. With the increase in the popularity of machine learn- ing (ML) and deep learning (DL) technologies, we are observing that an increasing number of HPC users 6 | are incorporating ML methods into their workflow and scientific discovery processes, across a wide spectrum of science domains such as biology, earth science, and physics. This gives rise to a diverse set of I/O patterns than the traditional checkpoint/restart-based HPC I/O behavior. The details of the I/O characteristics of such ML I/O workloads have not been studied extensively for large-scale leadership HPC systems. This paper aims to fill that gap by providing an in-depth analysis to gain an understanding of the I/O behavior of ML I/O workloads using darshan - an I/O characterization tool designed for lightweight tracing and profiling. We study the darshan logs of more than 23,000 HPC ML I/O jobs over a time period of one year running on Summit - the second-fastest supercomputer in the world. This paper provides a systematic I/O characterization of ML I/O jobs running on a leadership scale supercomputer to understand how the I/O behavior differs across science domains and the scale of workloads, and analyze the usage of parallel file system and burst buffer by ML I/O workloads. 7 | 8 | ## Bio 9 | Ahmad Maroof Karimi works as an HPC Operational Data Scientist in Analytics and A.I. Methods at Scale (AAIMS) Group in National Center for Computational Sciences (NCCS) Division, Oak Ridge National Laboratory. His current research focuses on the characterization of HPC I/O patterns and finding evolving HPC workload trends. He is also working on analyzing HPC facility data to characterize the HPC power consumption and building machine learning based job-aware power prediction models. Before joining ORNL, Ahmad completed his Ph.D. in Computer Science at CWRU, Cleveland, Ohio, in October 2020. His Ph.D. dissertation titled “Data science and machine learning to predict degradation and power of photovoltaic systems: convolutional and spatiotemporal graph neural networks” focused on classifying degradation mechanism and performance prediction of a photovoltaic power plant. He received his M.S. degree from the University of Toledo, Ohio, and B.S. degree from Aligarh Muslim University, India. Ahmad has also worked in the I.T. industry as a software programmer and database designer. 10 | -------------------------------------------------------------------------------- /abstracts/2022-08-09.md: -------------------------------------------------------------------------------- 1 | # Transparent Checkpointing: a mature technology enabling MANA for MPI and beyond 2 | ## Gene Cooperman, Khoury College of Computer Sciences, Northeastern University 3 | 4 | ## Abstract 5 | Although transparent checking grew up in the 1990s and 2000s as a technology for HPC, it has now grown as a tool that is useful for many newer domains. Today, this is no longer your grandfather's checkpointing software! In this talk, I will review some of the newer checkpointing technologies invented only in the last decade, and how they gate new capabilities that can be adapted in a variety of domains. 6 | This talk includes a tour of the 15-year old DMTCP project, with special emphasis on the latest achievement: MANA for MPI -- a robust package for transparent checkpointing of MPI. But as a prerequisite, one must have an understanding of two advances that brought DMTCP to its present state: (i) a general framework for extensible checkpointing plugins; and (ii) split processes (isolate the software application to be checkpointed from the underlying hardware). 7 | In the remainder of the talk, these two principles are first showcased in MANA. This is then followed by a selection of other domains where transparent checkpointing shows interesting potential. This includes: deep learning (especially for general frameworks), edge computing, lambda functions (serverless computing), spot instances, containers for parallel and distributed computing (Apptainer and Singularity), process migration (migrate the process to the data in joint work with JPL), deep debugging for parallel and distributed computations, a model for checkpointing in Hadoop, and more. 8 | 9 | ## Bio 10 | Professor Cooperman currently works in high-performance computing. He received his B.S. from the University of Michigan in 1974, and his Ph.D. from Brown University in 1978. He came to Northeastern University in 1986, and has been a full professor there since 1992. His visiting research positions include a 5-year IDEX Chair of Attractivity at the University of Toulouse/CNRS in France, and sabbaticals at Concordia University, at CERN, and in Inria/France. In 2014, he and his student, Xin Dong, used a novel idea to semi-automatically add multi-threading support to the million-line Geant4 code coordinated out of CERN. He is one of the more than 100 co-authors on the foundational Geant4 paper, whose current citation count is 34,000. Prof. Cooperman currently leads the DMTCP project (Distributed Multi-Threaded CheckPointing) for transparent checkpointing. The project began in 2004, and has benefited from a series of PhD theses. Over 150 refereed publications cite DMTCP as having contributed to their research project. 11 | -------------------------------------------------------------------------------- /abstracts/2019-02-15.md: -------------------------------------------------------------------------------- 1 | ## Nonlinear model reduction: Using machine learning to enable extreme-scale simulation for many-query problems 2 | ### Kevin Carlberg (Sandia National Laboratories) 3 | 4 | Physics-based modeling and simulation has become indispensable across many applications in engineering and science, ranging from nuclear-weapons design to monitoring national critical infrastructure. However, as simulation is playing an increasingly important role in scientific discovery, decision making, and design, greater demands are being placed on model fidelity. This high fidelity necessitates modeling fine spatiotemporal resolution, which can lead to extreme-scale, nonlinear dynamical-system models whose simulations consume months on thousands of computing cores. Further, most scientific applications (e.g., uncertainty quantification, design optimization) are many query in nature, as they require the (parameterized) model to be simulated thousands of times. This leads to a computational barrier: the computational cost of high-fidelity simulations renders them impractical for many-query problems. 5 | 6 | In this talk, I will present several advances in the field of nonlinear model reduction that exploit simulation data to overcome this barrier. These methods combine concepts from machine learning, computational mechanics, optimization, and goal-oriented adaptivity to produce low-dimensional reduced-order models (ROMs) that are 1) accurate, 2) low cost, 3) structure preserving, 4) reliable, and 5) certified. First, I will describe least-squares Petrov–Galerkin projection, which leverages subspace identification and optimal projection to ensure accuracy. Second, I will describe the sample mesh concept, which employs empirical regression and greedy-optimal sensor-placement techniques to ensure low cost. I will also describe novel methods that exploit time-domain data to further reduce computational costs. Third, I will present a technique that ensures the ROM is globally conservative in the case of finite-volume discretizations, thus ensuring structure preservation. Fourth, I will describe model reduction on nonlinear manifolds, wherein we employs convolutional autoencoders from deep learning to improve the predictive accuracy of the ROM. I will also describe ROM h-adaptivity, which employs concepts from adaptive mesh refinement to ensure that the ROM is reliable, i.e., it can satisfy any prescribed error tolerance. Finally, I will present machine-learning error models, which apply regression methods (e.g., feedforward neural networks) from machine learning to construct a stochastic model for the ROM error; this quantifies the ROM-induced epistemic uncertainty and provides a mechanism for certification. 7 | -------------------------------------------------------------------------------- /abstracts/2020-07-24.md: -------------------------------------------------------------------------------- 1 | # Making the invisible visible 2 | ## Srigokul Upadhyayula (Advanced Bioimaging Center, UC Berkeley) 3 | 4 | ## Abstract 5 | Direct physiological observation of subcellular dynamics is now feasible using the Lattice Light-Sheet Microscopy. It is a transformative imaging technology that spans the relevant scales in space and time because of the wide resolution range and large volumetric acquisition capability. This new imaging method allows us to record dynamics at a scale of nanometers and milliseconds, determine their consequences at a scale of microns and hours, and visualize their long-term outcome at a scale of up to several millimeters over several days. 6 | 7 | I’ll present our past work on combined lattice light sheet microscopy with adaptive optics to achieve (Science, 2018), across large multicellular volumes, noninvasive aberration-free imaging of subcellular dynamics in vivo. Next, I’ll discuss the combination of lattice light-sheet with the physical expansion of samples (Expansion Microscopy) that enables scalable super-resolution volumetric imaging of large tissues (Science, 2019) including the complete fly brain, columns of mouse brain – datasets spanning several hundred terabytes. Finally, I will introduce our next-generation microscope design– dubbed the “Swiss army knife microscope”, which combines at least ten different modes of imaging with integrated light paths. In essence, this new microscope is designed to seamlessly switch between modes of imaging in order to alleviate the tradeoffs related to resolution, speed, invasiveness and imaging depth, which precludes any single optical microscopy to function optimally for a diverse set of biological specimens. 8 | 9 | ## Bio 10 | Gokul Upadhyayula uses applied engineering and basic science to build cutting-edge, adaptive optical multi-functional microscopes to enable imaging across scales spanning several orders of magnitude in space and time. As the leading scientific director at UC Berkeley’s Advanced BioImaging Center (ABC), Gokul's vision is to provide state-of-the-art microscopy, and dedicated human and hardware resources capable of handling terabyte to petabyte scale projects. In addition to this, his mission is to develop robust, open source computational workflows that allow scientists to extract biologically meaningful insights. His earlier work includes studying the charge transfer properties of cyanine dyes and bio-inspired electrets using ultra-fast femtosecond spectroscopy during his time at University of California, Riverside; and using lattice light-sheet microscopy (LLSM) with high temporal and spatial resolution at the molecular level during his postdoctoral research at Harvard Medical School / Boston Children’s Hospital. 11 | -------------------------------------------------------------------------------- /abstracts/2021-11-30.md: -------------------------------------------------------------------------------- 1 | # Tiled: A Service for Structured Data Access 2 | ## Dan Allan (NSLS-II - Brookhaven National Laboratory) 3 | 4 | ## Abstract 5 | In the Data Science and Systems Integration Program at NSLS-II, we have explored various ways to separate I/O code from user science code. After seven years of developing in-house solutions and contributing to external ones (including Intake), we propose an abstraction that we think is a broadly useful building block, named Tiled. Tiled is a data access service for data-aware portals and data science tools. It has a Python client that feels much like h5py to use and integrates naturally with dask, but nothing about the service is Python-specific; it also works from curl. Tiled’s service sits atop databases, filesystems, and/or remote services to enable search and structured, chunk-wise access to data in an extensible variety of appropriate formats, providing data in a consistent structure regardless of the format the data happens to be stored in at rest. The natively-supported formats span slow but widespread interchange formats (e.g. CSV, JSON) and fast, efficient ones (e.g. C buffers, Apache Arrow Tables). Tiled enables slicing and sub-selection to read and transfer only the data of interest, and it enables parallelized download of many chunks at once. Users can access data with very light software dependencies and fast partial downloads. Tiled puts an emphasis on structures rather than formats, including N-dimensional strided arrays (i.e. numpy-like arrays), tabular data (i.e. pandas-like“dataframes”), and hierarchical structures thereof (e.g. xarrays, HDF5-compatible structures like NeXus). Tiled implements extensible access control enforcement based on web security standards, similar -to JupyterHub Authenticators. Like Jupyter, Tiled can be used by a single user or deployed as a shared resource. Tiled facilitates local client-side caching in a standard web browser or in Tiled’s Python client, making efficient use of bandwidth and enabling an offline “airplane mode.” Service-side caching of "hot" datasets and resources is also possible. Tiled is conceptually “complete” but still new enough that there is room for disruptive suggestions and feedback. We are interested in particular in exploring how Tiled could be made broadly available to NERSC users alongside traditional file-based access, and how that work might prompt us to rethink aspects of Tiled’s design. 6 | 7 | ## Bio 8 | Dan Allan is scientific software developer and group lead in the Data Science and Systems Integration Program at NSLS-II. He joined Brookhaven National Lab as a post-doc in 2015 after studying soft condensed-matter experimental physics and getting involved in the open source scientific Python community. He works on data acquisition, management, and analysis within and around the "Bluesky" software ecosystem. 9 | -------------------------------------------------------------------------------- /abstracts/2022-02-01.md: -------------------------------------------------------------------------------- 1 | # KubeFlux: a scheduler plugin bridging the cloud-HPC gap in Kubernetes 2 | ## Claudia Misale (IBM Research), Daniel J. Milroy (Lawrence Livermore National Laboratory) 3 | 4 | ## Abstract 5 | The cloud is an increasingly important market sector of computing and is driving innovation. Adoption of cloud technologies by high performance computing (HPC) is accelerating, and HPC users want their applications to perform well everywhere. While cloud orchestration frameworks like Kubernetes provide advantages like resiliency, elasticity, and automation, they are not designed to enable application performance to the same degree as HPC workload managers and schedulers. As HPC and cloud Computing converge, techniques from HPC can be integrated into the cloud to improve application performance and provide universal scalability. We present KubeFlux, a Kubernetes plugin based on the Fluxion open-source HPC scheduler component of the Flux framework developed at the Lawrence Livermore National Laboratory. We introduce the Flux framework and the Fluxion scheduler and describe how their hierarchical, graph-based foundation is naturally suited to converged computing. We discuss uses for KubeFlux and compare the performance of an application scheduled by the Kubernetes default scheduler and KubeFlux. KubeFlux is an example of the rich capability that can be added to Kubernetes and paves the way to democratization of the cloud for HPC workloads. 6 | 7 | ## Bios 8 | Claudia Misale is a Research Staff Member in the Hybrid Cloud Infrastructure Software group at IBM T.J. Watson Research Center (NY). Her research is focused on Kubernetes for IBM Public Cloud, and also targets porting HPC applications to the cloud by enabling batch scheduling alternatives for Kubernetes. She is mainly interested in cloud computing and container technologies, and her background is on high-level parallel programming models and patterns, and big data analytics on HPC platforms. She received her master’s summa cum laude and bachelor’s degree in Computer Science at the University of Calabria (Italy), and her PhD from the Computer Science Department of the University of Torino (Italy). 9 | 10 | Daniel Milroy is a Computer Scientist at the Center for Applied Scientific Computing at the Lawrence Livermore National Laboratory. His research focuses on graph-based scheduling and resource representation and management for high performance computing (HPC) and cloud converged environments. While Dan’s research background is numerical analysis and software quality assurance and correctness for climate simulations, he is currently interested in scheduling and representing dynamic resources, and co-scheduling and management techniques for HPC and cloud. Dan holds a B.A. in physics from the University of Chicago, and an M.S. and PhD in computer science from the University of Colorado Boulder. 11 | -------------------------------------------------------------------------------- /abstracts/2020-02-14.md: -------------------------------------------------------------------------------- 1 | # Intrinsiccomputation and physics-based machine learning for emergentself-organization in far-from-equilibrium systems 2 | ## Adam Rupe (UC Davis) 3 | 4 | ## Abstract 5 | Coherent structures formspontaneously in far-from-equilibrium spatiotemporal systems and are foundat all spatial scales in natural phenomena from laboratory hydrodynamicflows and chemical reactions to ocean and atmosphere dynamics.Phenomenologically, they appear as key components that organize macroscopicdynamical behaviors. Unlike their equilibrium and near-equilibriumcounterparts, there is no general theory to predict what patterns andstructures may emerge in far-from-equilibrium systems. Each system behavesdifferently; details and history matter. The complex behaviors that emergecannot be explicitly described mathematically, nor can they be directlydeduced from the governing equations (e.g. what is the mathematicalexpression for a hurricane, and how can you derive it from the equations ofa general circulation climate model?). It is thus appealing to bring theinstance-based data-driven models of machine learning to bear on theproblem. Supervised learning models have been the most successful, but theyrequire ground-truth training labels which do not exist forfar-from-equilibrium structures. Unsupervised models that leverage physicalprinciples of self-organization are required. To this end we will makeconnections between structural organization and intrinsic computation tomotivate the use of physics-based unsupervised models called local causalstates. As local models they are capable of capturing structures ofarbitrary shape and size in a visually interpretable manner, due to theshared coordinate geometry between observable spacetime fields and theirassociated latent local causal state fields. We will show the local causalstates can capture patterns in cellular automata models as generalizedspacetime symmetries and coherent structures as localized deviations fromthese generalized symmetries. To demonstrate their applicability toreal-world systems, we show the utility of the local causal states forextracting coherent structures in simulations and observations of complexfluid flows, including promising results highlighting extreme weatherevents in the water vapor field of the CAM5.1 climate model. These resultsrequire high-performance computing, and we will briefly describe how wewere able to process almost 90TB in under 7 minutes end-to-end on 1024Haswell nodes of Cori using a distributed implementation in Python. 6 | 7 | 8 | ## Bio 9 | Adam Rupe is a PhD candidate atthe University of California Davis. His research interests center oncomplex dynamical systems, including theoretical aspects and data-drivenapproaches for applications to real-world problems. He was lead author onProject DisCo, which was chosen for an HPC Innovation and Excellent Awardby HPC User Forum and Hyperion Research. 10 | -------------------------------------------------------------------------------- /abstracts/2021-04-16.md: -------------------------------------------------------------------------------- 1 | # The Open Catalyst 2020 (OC20) Dataset and Community Challenges 2 | ## Zachary Ulissi (Carnegie Mellon University),
Larry Zitnick (Facebook AI Research) 3 | 4 | ## Abstract 5 | The Open Catalyst Project aims to develop new ML methods and models to accelerate the catalyst 6 | simulation process for renewable energy technologies and improve our ability to predict 7 | activity/selectivity across catalyst composition. To achieve that in the short term we need 8 | participation from the ML community in solving key challenges in catalysis. One path to 9 | interaction is the development of grand challenge datasets that are representative of common 10 | challenges in catalysis, large enough to excite the ML community, and large enough to take 11 | advantage of and encourage advances in deep learning models. Similar datasets have had a large 12 | impact in small molecule drug discovery, organic photovoltaics, and inorganic crystal structure 13 | prediction. We present the first open dataset from this effort on thermochemical intermediates 14 | across stable multi-metallic and p-block doped surfaces. This dataset includes full-accuracy DFT 15 | calculations across 53 elements and their binary/ternary materials, various low-index facets. 16 | Adsorbates span 56 common reaction intermediates with relevance to carbon, oxygen, and nitrogen 17 | thermal and electrochemical reactions. Off-equilibrium structures are also generated and included 18 | to aid in machine learning force field design and fitting. Collectively, this dataset represents 19 | the largest systematic dataset that bridges organic and inorganic chemistry and will enable a new 20 | generation of catalyst structure/property relationships. Fixed train/test splits that represent 21 | common chemical challenges and an open challenge website will be discussed to encourage 22 | competition and buy-in from the ML community. 23 | 24 | ## Bio 25 | Zachary Ulissi is an Assistant Professor of Chemical Engineering at Carnegie Mellon University. He 26 | works on the development and application of high-throughput computational methods in catalysis, 27 | machine learning models to predict their properties, and active learning methods to guide these 28 | systems. Applications include energy materials, CO2 utilization, fuel cell development, and 29 | additive manufacturing. He has been recognized nationally for his work including the 3M 30 | Non-Tenured Faculty Award and the AIChE 35-under-35 award among others. 31 | 32 | Larry Zitnick is a research scientist at Facebook AI Research in Menlo Park. His current areas of 33 | interest include scientific applications of AI, language and vision, and object recognition. He 34 | serves on the board of the Common Visual Data Foundation whose mission is to aid the computer 35 | vision community in creating datasets and competitions. Previously, he spent 12 great years at 36 | Microsoft Research, and obtained a PhD in Robotics from CMU's Robotics Institute. 37 | -------------------------------------------------------------------------------- /abstracts/2020-09-04.md: -------------------------------------------------------------------------------- 1 | # Steps toward holistic control of particle accelerators with neural networks 2 | ## Auralee Edelen (SLAC National Accelerator Laboratory) 3 | 4 | ## Abstract 5 | Particle accelerators are used in a wide array of medical, industrial, and scientific applications, ranging from cancer treatment to understanding fundamental laws of physics. While each of these applications brings with them different operational requirements, a common challenge concerns how to optimally adjust controllable settings of the accelerator to obtain the desired beam characteristics. For example, at highly flexible user facilities like the LCLS and FACET-II, requests for a wide array custom beam configurations must be met in a limited window of time to ensure the success of each experiment — a task which can be difficult both in terms of tuning time and the final achievable solution quality, especially for novel or non-standard setups. At present, the operation of most accelerator facilities relies heavily on manual tuning by highly-skilled human operators, sometimes with the aid of simplified physics models and local optimization algorithms. As a complement to these existing tools, approaches based on machine learning are poised to enhance our ability to achieve higher-quality beams, fulfill requests for custom beam parameters more quickly, and aid the development of novel operating schemes. Focusing on neural network based approaches, I will discuss proof-of-principle studies that point toward the potential of machine learning in this regard, highlight open questions and challenges, and give an outlook on some of the future pathways toward bringing these techniques more fully into operation of accelerators. These improvements could increase the scientific output of user facilities and enable new capabilities by tuning a wider range of machine settings, as well as exploiting subtle sensitivities that may otherwise go unutilized. They could also help us to meet the modeling and tuning challenges that become more acute as we push toward the more difficult-to-achieve beam parameters that are desired for future accelerator applications (e.g. higher beam energies and intensities, higher stability, and extreme adjustments of the beam shape in phase space). 6 | 7 | 8 | ## Bio 9 | Auralee is a Panofsky Fellow at SLAC National Accelerator Laboratory, where she works on developing machine learning based approaches for modeling and control of particle accelerators. She arrived at SLAC as a research associate in 2018. During her graduate studies, Auralee worked with Fermi National Accelerator Laboratory on early proof-of-principle studies in applying modern neural network based approaches to particle accelerators. Auralee also has been active in the particle accelerator community with regard to education and promotion of machine learning, including, for example, helping to organize and provide tutorials for several workshops on machine learning for particle accelerator applications. 10 | -------------------------------------------------------------------------------- /abstracts/2022-09-13.md: -------------------------------------------------------------------------------- 1 | # Pegasus Workflow Management System 2 | ## Karan Vahi, Information Sciences Institute, University of Southern California 3 | 4 | ## Abstract 5 | Workflows are a key technology for enabling complex scientific computations. They capture the interdependencies between processing steps in data analysis and simulation pipelines as well as the mechanisms to execute those steps reliably and efficiently. Workflows can capture complex processes to promote sharing and reuse, and also provide provenance information necessary for the verification of scientific results and scientific reproducibility. Pegasus (https://pegasus.isi.edu) is being used in a number of scientific domains doing production grade science. In 2016 the LIGO gravitational wave experiment used Pegasus to analyze instrumental data and confirm the first detection of a gravitational wave. The Southern California Earthquake Center (SCEC) based at USC, uses a Pegasus managed workflow infrastructure called Cybershake to generate hazard maps for the Southern California region. In 2021, SCEC conducted a CyberShake study on DOE systems Summit that used a simulation-based ERF for the first time. Overall, the study required 65,470 node-hours (358,000 GPU-hours and 243,000 CPU-hours ) of computation with Pegasus submitting tens of thousands of remote jobs automatically, and managed 165 TB of data over the 29-day study. Pegasus is also being used in astronomy, bioinformatics, civil engineering, climate modeling, earthquake science, molecular dynamics and other complex analyses. Pegasus users express their workflows using an abstract representation devoid of resource- specific information. Pegasus plans these abstract workflows by mapping tasks to available resources, augmenting the workflow with data management tasks, and optimizing the workflow by grouping small tasks into more efficient clustered batch jobs. Pegasus then executes this plan. If an error occurs at runtime, Pegasus automatically retries the failed task and provides checkpointing in case the workflow cannot continue. Pegasus can record provenance about the data, software and hardware used. Pegasus has a foundation for managing workflows in different environments, using workflow engines that are customized for a particular workload and system. Pegasus has a well defined support for major container technologies such as Docker, Singulartiy, Shifter that allows users to have the jobs in their workflow use containers of their choice. Pegasus most recent major release Pegasus 5.0 is a major improvement over previous releases. Pegasus 5.0 provides a brand new Python3 workflow API developed from the ground up so that, in addition to generating the abstract workflow and all the catalogs, it now allows you to plan, submit, monitor, analyze and generate statistics of your workflow. 6 | 7 | ## Bio 8 | Karan Vahi is a Senior Computer Scientist in the Science Automation Technologies group at the USC Information Sciences Institute. He has been working in the field of scientific workflows since 2002, and has been closely involved in the development of the Pegasus Workflow Management System. He is currently the architect/lead developer for Pegasus and in charge of the core development of Pegasus. His work on implementing integrity checking in Pegasus for scientific workflows won the best paper and the Phil Andrews Most Transformative Research Award at PEARC19. He currently leads the Cloud Platforms group at CI Compass, a NSF CI Center, which includes CI practitioners from various NFS Major facilities(MF’s) and aims to understand the current practices for Cloud Infrastructure used by MFs and research alternative solutions. https://www.isi.edu/directory/vahi/ 9 | -------------------------------------------------------------------------------- /abstracts/2022-03-15.md: -------------------------------------------------------------------------------- 1 | # Chameleon: An Innovation Platform for Computer Science Research and Education 2 | ## Kate Keahey, Argonne National Lab, UChicago CASE 3 | 4 | ## Abstract 5 | We live in interesting times: new ideas and technological opportunities emerge at ever increasing rate in disaggregated hardware, programmable networks, and the edge computing and IoT space to name just a few. These innovations require an instrument where they can be deployed and investigated, and where new solutions that those disruptive ideas require can be developed, tested, and shared. To support a breadth of Computer Science experiments such instrument has to provide access to a diversity of hardware configurations, support deployment at scale, as well as deep reconfigrability so that a wide range of experiments can be supported. It also has to provide mechanisms for easy and direct sharing of repeatable digital artifacts so that new experiments and results can be easily replicated and help enable further innovation. Most importantly -- since science does not stand still – such instrument requires the capability for constant adaptation to support an ever increasing range of experiments driven by emergent ideas and opportunities. The NSF-funded Chameleon testbed (www.chameleoncloud.org) has been developed to provide all those capabilities. It provides access to a variety of hardware including cutting-edge architectures, a range of accelerators, storage hierarchies with a mix of large RAM, NVDIMMs, a variety of enterprise and consumer grade SDDs, HDDs, high-bandwidth I/0 storage, SDN-enabled networking hardware, and fast interconnects. This diversity was enlarged recently to add support for edge computing/IoT devices and will be further extended this year to include LiQid composable hardware as well as P4 switches. Chameleon is distributed over two core sites at the University of Chicago and the Texas Advanced Computing Center (TACC) connected by 100 Gbps network – as well as three volunteer sites at NCAR, Northwestern University, and the University of Illinois in Chicago (UIC). Bare metal reconfigurability for Computer Science experiments is provided by CHameleon Infrastructure (CHI), based on an enhanced bare-metal flavor of OpenStack: it allows users to reconfigure resources at bare metal level, boot from custom kernel, and have root privileges on the machines. To date, the testbed has supported 6,000+ users and 800+ projects in research, education, and emergent applications. In this talk, I will describe the goals, the design strategy, and the capabilities of the testbed, as well as some of the research and education projects our users are working on. I will also discuss our new thrusts in support for research on edge computing and IoT, our investment in developing and packaging of research infrastructure (CHI-in-a-Box), as well as our support for composable systems that can both dynamically integrate resources from other sources into Chameleon and make Chameleon resources available via other systems. Lastly, I will describe the services and tools we created to support sharing of experiments, educational curricula, and other digitally expressed artifacts that allow science to be shared via active involvement and foster reproducibility. 6 | 7 | ## Bio 8 | Kate Keahey is one of the pioneers of infrastructure cloud computing. She created the Nimbus project, recognized as the first open source Infrastructure-as-a-Service implementation, and continues to work on research aligning cloud computing concepts with the needs of scientific datacenters and applications. To facilitate such research for the community at large, Kate leads the Chameleon project, providing a deeply reconfigurable, large-scale, and open experimental platform for Computer Science research. To foster the recognition of contributions to science made by software projects, Kate co-founded and serves as co-Editor-in-Chief of the SoftwareX journal, a new format designed to publish software contributions. Kate is a Scientist at Argonne National Laboratory and a Senior Scientist The University of Chicago Consortium for Advanced Science and Engineering (UChicago CASE). 9 | -------------------------------------------------------------------------------- /2019.md: -------------------------------------------------------------------------------- 1 | ### 2019 Data Seminars 2 | |Date |Title |Speaker |Host |Slides | 3 | |-----|---------------------|------------------------|-------------------|-------------| 4 | |1/25 |Hierarchical Deep Learning for Long-term Sequence Generation ([abstract](abstracts/2019-01-25.md))|Stephan Zheng (Salesforce Research) |Steven Farrell |[keynote][2], [pdf][1]| 5 | |2/1 |Mesh-TensorFlow: Deep Learning for Supercomputers ([abstract](abstracts/2019-02-01.md))|Noam Shazeer (Google Brain) |Mustafa Mustafa | | 6 | |2/8 |An Empirical Model of Large-Batch Training ([abstract](abstracts/2019-02-08.md))|Sam McCandlish (OpenAI) |Mustafa Mustafa | | 7 | |2/15 |Nonlinear model reduction: Using machine learning to enable extreme-scale simulation for many-query problems ([abstract](abstracts/2019-02-15.md))|Kevin Carlberg (Sandia Natl. Labs)|Karthik Kashinath | | 8 | |2/22 |Learning quantum states with generative models ([abstract](abstracts/2019-02-22.md))|Juan Carrasquilla (Vector Institute) |Karthik Kashinath | | 9 | |3/1 |Jupyter at NERSC ([abstract](abstracts/2019-03-01.md))|Rollin Thomas (NERSC, LBL) |Prabhat | | 10 | |3/8 |Introduction to Deep Learning ([abstract](abstracts/2019-03-08.md))|Mustafa Mustafa (NERSC, LBL) |Prabhat |[pdf][3] | 11 | |3/15 |SENSE: SDN for End-to-end Networked Science at the Exascale ([abstract](abstracts/2019-03-15.md))|Chin Guok(ESNet, LBL) |David Skinner | | 12 | |3/22 |Spatio-temporal modeling using ML ([abstract](abstracts/2019-03-22.md))|Rose Yu (NorthEastern Univ.) |Karthik Kashinath, Adrian Albert | | 13 | |3/29 |GANs for Soil Mechanics ([abstract](abstracts/2019-03-29.md))|Utkarsh Mital (Caltech) |Adrian Albert | | 14 | |4/12 |Sizing Neural Network Experiments ([abstract](abstracts/2019-04-12.md))|Gerald Friedland (UCB & LLNL) |Aydin Buluc | | 15 | |4/19 |Picture Perfect ([abstract](abstracts/2019-04-19.md))|Peter Denes (LBL) |David Skinner | | 16 | |5/3 |Infusing Structure into Machine Learning Algorithms ([abstract](abstracts/2019-05-03.md))|Animashree Anandkumar (Caltech, NVIDIA)|Karthik Kashinath | | 17 | |5/21 |Cascade Reconstruction in IceCube using Convolutional and Generative Neural Networks ([abstract](abstracts/2019-05-21.md))|Mirco Hunnefeld (TU Dortmund)|Lisa Gerhardt | | 18 | |5/24 |Maglev and the Future of Long Distance Transportation ([abstract](abstracts/2019-05-24.md))|John van Rosendale (College of William and Mary)|Prabhat | | 19 | |5/31 |Reflections on Human Space Flight” subtitled “Why Single Planet Species Don’t Survive) ([abstract](abstracts/2019-05-31.md))|Jim Newman (Naval Postgraduate School)|Prabhat | | 20 | |6/14 |Optimizing Graph Algorithms |Shaikh Arifuzzman |Prabhat | | 21 | |7/26 |W3C and the Future of Data Sharing on the Web |Annette Greiner | | | 22 | |8/9 |Learning for HPC Systems: Progress and Challenges |Taylor Groves | | | 23 | |8/23 |Physics-informed Autoencoders for Lyapunov-stable Fluid Flow Prediction |N. Benjamin Erichson |John Wu / Jialin Liu|| 24 | |8/30 |Accelerating Deep Learning with FPGAs |Rahul Namiyar |Prabhat | | 25 | |10/4 |Update on the Cerebras AI Accelerator |Andy Hock & Jessical Liu |John Shalf | | 26 | |10/25 |Opportunities and Challenges in Linking DAQ & HPC Systems ([abstract](abstracts/2019-10-25.md)) |David Skinner |Prabhat ||| 27 | |11/1|Machine Learning, Synthetic Biology and Automation: Engineering Life for the Benefit of Society ([abstract](abstracts/2019-11-01.md))|Hector Garcia Martin |Steven Farrell |[ppt][4]| 28 | |11/15|FlowPM: Particle-Mesh N-body Simulation in TensorFlow ([abstract](abstracts/2019-11-15.md))|Chirag Modi |Mustafa Mustafa |[pdf][6]| 29 | |12/6 |Natural Language Processing for Materials Discovery and Design ([abstract](abstracts/2019-12-06.md))|John Dagdelen |Steven Farrell |[pdf][5]| 30 | 31 | [1]: https://drive.google.com/open?id=1uTJjAPPnvY4ds0_02_jeYP9Uh3NyX8KG 32 | [2]: https://drive.google.com/open?id=141SLMMx1mmJp3ZssruJ_AIR7K9OtSAPh 33 | [3]: https://docs.google.com/presentation/d/1haI_h9jbvcSyM_ngCSC9-ZknLBJ6JDqVlb7NXk4NA3Y/edit#slide=id.g4c14c04cdd_0_111 34 | [4]: https://drive.google.com/file/d/1WxAev_SDQ0EY5awKWJjrDiCvlmx0au4n/view?usp=sharing 35 | [5]: https://drive.google.com/file/d/1BAtOS6cO6vuMw9TGLpuesKIQAhUCbo66/view?usp=sharing 36 | [6]: https://modichirag.github.io/talks/LBNL2019/#/ 37 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | ## NERSC Data Seminars Series - Lawrence Berkeley National Laboratory 2 | 3 | The NERSC Data Seminar Series are held at [Berkeley Lab](https://www.lbl.gov/). The series hosts speakers to: 4 | - Learn about latest science and methods results from researchers 5 | - Learn from software vendors on their product offerings 6 | - Facilitate communications between NERSC and other lab CS staff 7 | 8 | ### Time: 9 | 10 | Talks are held at **11am-12pm** on **Tuesdays** and are posted on the [CS Seminars Calendar](https://www.nersc.gov/events/cs-seminars/). 11 | If you are affiliated with Berkeley Lab you can sign up to receive announcements about the Machine Learning seminars at the [ML4Sci mailing-list](https://groups.google.com/a/lbl.gov/forum/#!forum/ml4sci). 12 | 13 | ### Remote attendance: 14 | 15 | The Zoom link will be visible in the calendar invite. 16 | 17 | ### Contacting the speakers: 18 | 19 | Feel free to contact the host with questions or requests for time with the speaker. 20 | 21 | ### Videos: 22 | 23 | Video recordings are available in the ["Data Seminars Series" playlist](https://www.youtube.com/playlist?list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy) on YouTube. 24 | 25 | ### Past years: 26 | 27 | - [2021](2021.md) 28 | - [2020](2020.md) 29 | - [2019](2019.md) 30 | 31 | ### 2022 Seminars 32 | 33 | |Date |Title |Speaker |Host |Material | 34 | |-----|---------------------|------------------------|-------------------|---------------| 35 | |1/13 |The Next Generation of High Performance Computing: HPC-2.0 ([abstract](abstracts/2022-01-13.md))|Gregory Kurtzer (Ctrl IQ, Rocky Linux)|Jonathan Skone|[video][1], [slides][2]| 36 | |2/1 |KubeFlux: a scheduler plugin bridging the cloud-HPC gap in Kubernetes ([abstract](abstracts/2022-02-01.md))|Claudia Misale (IBM), Daniel J. Milroy (LLNL)|Shane Canon|[video][3], [slides][4]| 37 | |3/15 |Chameleon: An Innovation Platform for Computer Science Research and Education ([abstract](abstracts/2022-03-15.md))|Kate Keahey (ANL, UChicago CASE)|Shane Canon, Jonathan Skone|[video][5], [slides][6]| 38 | |3/22 |Composable Platforms for Scientific Computing: Experiences and Outcomes ([abstract](abstracts/2022-03-22.md))|Brian Werts, Sam Weekly and Erik Gough (Purdue University)|Jonathan Skone|[video][7], [slides][8]| 39 | |4/12 ||||| 40 | |4/19 |Discovering and Modeling Strong Gravitational Lenses with Cori and Perlmutter at NERSC ([abstract](abstracts/2022-04-19.md))|Xiaosheng Huang (USF), Andi Gu (UCB)|Steve Farrell|[video][9]| 41 | |4/26 |FirecREST, RESTful HPC ([abstract](abstracts/2022-04-26.md))|Juan Pablo Dorsch (CSCS)|Jonathan Skone|[video][10], [slides][11]| 42 | |5/17 |Memory Disaggregation: Potentials and Pitfalls ([abstract](abstracts/2022-05-17.md))|Nan Ding (LBL)|Hai Ah Nam|[video][12]| 43 | |5/19 |Quantum Computing for NERSC Staff ([abstract](abstracts/2022-05-19.md))|Katie Klymko, Daan Camps & Jan Balewski (NERSC/LBL)|Katie Klymko, Daan Camps, Jan Balewski|[video][13]| 44 | |6/7 |Building a Platform for Operating Multi-Institutional Distributed Services ([abstract](abstracts/2022-06-07.md))|Lincoln Bryant (Enrico Fermi Institute - University of Chicago)|Jonathan Skone|[video][14], [slides][15]| 45 | |6/14 |Artificial Design of Porous Materials ([abstract](abstracts/2022-06-14.md))|Jihan Kim (KAIST)|Brian Austin|[video][16]| 46 | |6/23 |Demo and hands-on session on ReFrame ([abstract](abstracts/2022-06-23.md))|Lisa Gerhardt, Alberto Chiusole|Lisa Gerhardt, Alberto Chiusole|[video][17], [slides][18]| 47 | |6/28 |FourCastNet: Data-driven, high-resolution atmosphere modeling at scale ([abstract](abstracts/2022-06-28.md))|Shashank Subramanian (NERSC/LBL)|Peter Harrington|[video][19], [slides][20]| 48 | |8/9 |Transparent Checkpointing: a mature technology enabling MANA for MPI and beyond ([abstract](abstracts/2022-08-09.md))|Gene Cooperman, Khoury College of Computer Sciences, Northeastern University|Zhengji Zhao|[video][21], [slides][22]| 49 | |9/13 |Pegasus Workflow Management System ([abstract](abstracts/2022-09-13.md))|Karan Vahi, Information Sciences Institute, University of Southern California|Hai Ah Nam|[video][23], [slides][24]| 50 | |X/X |TBD | Chris Mungall (LBL) | Shane Canon | | 51 | 52 | [1]: https://www.youtube.com/watch?v=isP-Hqw_-nc&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 53 | [2]: https://drive.google.com/file/d/1knKvXpL1bghN5U0zpOWgpsmBp7y370aJ/view?usp=sharing 54 | [3]: https://www.youtube.com/watch?v=RSsuamxKxH0&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 55 | [4]: https://drive.google.com/file/d/1am44USKAM2GBcXo381fF0q4TUNngtHLU/view?usp=sharing 56 | [5]: https://www.youtube.com/watch?v=fMWdmEIZldc&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 57 | [6]: https://drive.google.com/file/d/1pkX9dtsiPA6k_2v6WpRGRYDUBeijtjq-/view?usp=sharing 58 | [7]: https://www.youtube.com/watch?v=5patk3AVBoY&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 59 | [8]: https://drive.google.com/file/d/17DMO4fCBoQ0e3pKGq_jhOZ_vzXWHClC_/view?usp=sharing 60 | [9]: https://www.youtube.com/watch?v=BdpgOhxEyMs&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 61 | [10]: https://www.youtube.com/watch?v=9O1L4Wf9sZs&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 62 | [11]: https://drive.google.com/file/d/10UkMJUILL4J4yAIbcOv4yG5QgS7hgx0p/view?usp=sharing 63 | [12]: https://www.youtube.com/watch?v=5p98UeSIsSY&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 64 | [13]: https://www.youtube.com/watch?v=TyA8XwwJ350&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 65 | [14]: https://www.youtube.com/watch?v=253Z5M0Ps_A&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 66 | [15]: https://drive.google.com/file/d/1PXuxQXmcGyZQjhtNtzlfzOfIvIWJ-5ck/view?usp=sharing 67 | [16]: https://www.youtube.com/watch?v=X7cW-HGTdQw&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 68 | [17]: https://www.youtube.com/watch?v=PC-EzR_f1cA&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 69 | [18]: https://drive.google.com/file/d/15C5IcFcm1_KHTThKwOC8o50sC3dl2YMD/view?usp=sharing 70 | [19]: https://www.youtube.com/watch?v=RIUV1vpNalI&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 71 | [20]: https://drive.google.com/file/d/1jJwTZnwZ2ieEQABbFaUvv-lIUfRFjNGW/view?usp=sharing 72 | [21]: https://www.youtube.com/watch?v=e6MlEXZD3mQ&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 73 | [22]: https://drive.google.com/file/d/10PnYYnYGz1d7D4CuOQxzUAP2lzAXNeL7/view?usp=sharing 74 | [23]: https://www.youtube.com/watch?v=l2got-jtqEU&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 75 | [24]: https://drive.google.com/file/d/187lS-uxNp1ny1T-K05Tkf_XwAAWKdnNb/view?usp=sharing 76 | -------------------------------------------------------------------------------- /2021.md: -------------------------------------------------------------------------------- 1 | ### 2021 Seminars 2 | 3 | |Date |Title |Speaker |Host |Material | 4 | |-----|---------------------|------------------------|-------------------|---------------| 5 | |1/12 |Deep Learning Approaches for Modeling Multi-Scale Chaos and Geophysical Turbulence ([abstract](abstracts/2021-01-12.md))|Ashesh Chattopadhyay (Rice University/LBL)|Mustafa Mustafa|[video][3]| 6 | |1/19 |Self-Supervised Representation Learning for Astronomical Images ([abstract](abstracts/2021-01-19.md))|Md. Abul Hayat (UARK/LBL),
George Stein (UCB/LBL)|Mustafa Mustafa|[pdf][1], [key][2], [video][4]| 7 | |2/2 |Machine learning as a tool for Standard Model measurements ([abstract](abstracts/2021-02-02.md))|Vinicius Mikuni (Univ. of Zurich)|Mustafa Mustafa|| 8 | |2/23 |The CS Area Superfacility Project: Year in Review 2020 |The Superfacility Team|Debbie Bard|[video][6], [pdf][5]| 9 | |3/9 |Darshan: Enabling Application I/O Understanding in an Evolving HPC Landscape ([abstract](abstracts/2021-03-09.md))|Shane Snyder (Argonne National Laboratory)|Alberto Chiusole|[video][7]| 10 | |4/16 |The Open Catalyst 2020 (OC20) Dataset & Community Challenges ([abstract](abstracts/2021-04-16.md))|Zachary Ulissi (Carnegie Mellon University),
Larry Zitnick (Facebook AI Research)|Brandon Wood|[video][8]| 11 | |6/29 | FAIR Data - What's all the fuss about?|Heather Coates (IUPUI)| Kristy Kallback-Rose| [pdf][9], [video][10]| 12 | |7/13 |Monitoring Scientific Python Usage at NERSC ([abstract](abstracts/2021-07-13.md))|Rollin Thomas (NERSC/LBL)|Wahid Bhimji|[video][11]| 13 | |8/3 |Legate: High Productivity High Performance Computing ([abstract](abstracts/2021-08-03.md))|Manolis Papadakis (NVIDIA)|Laurie Stephey|[video][12], [pdf][13]| 14 | |8/10 |LDMS data at NERSC: doing something useful with 8 PB of CSV files ([abstract](abstracts/2021-08-10.md))|Brandon Cook (NERSC/LBL)|Wahid Bhimji|[video][14], [pdf][15]| 15 | |8/24 |JGI Computing - the future is looking cloudy ([abstract](abstracts/2021-08-24.md))|Kjiersten Fagnan (JGI/LBL)|Nicholas Wright|[video][16]| 16 | |9/14 |Challenges and successes with a hybrid multicloud implementation for research computing ([abstract](abstracts/2021-09-14.md))|Jonathan Skone (NERSC/LBL)|Nicholas Wright|[video][17]| 17 | |9/21 |Darshan 3.3.1 & Autoperf 2.0 updates ([abstract](abstracts/2021-09-21.md))|Sudheer Chunduri (Argonne National Lab), Kevin Harms (Argonne National Lab)|Taylor Groves|[video][18]| 18 | |9/28 |Fusion Long Range Plan and Fusion Energy Sciences Advisory Committee Report Briefing and Current and Future FES Needs at NERSC ([abstract](abstracts/2021-09-28.md))|Richard Hawryluk (PPPL), Troy Carter (UCLA), Brian Wirth (ORNL), Chris Holland (UCSD), Dave Humphreys (General Atomics)|Richard Gerber|[video][19]| 19 | |10/19|Scaling out HPC with On-Premise Performance in the Oracle Cloud Infrastructure ([abstract](abstracts/2021-10-19.md))|Luiz DeRose, Ph.D. (Oracle)|Jonathan Skone|[video][20]| 20 | |10/26|Challenges and Directions in ML System Performance: The MLPerf Story ([abstract](abstracts/2021-10-26.md))|David Kanter (MLCommons)|Hai Ah Nam|[video][21]| 21 | |11/2 |Characterizing I/O behavior of large-scale scientific Deep Learning applications ([abstract](abstracts/2021-11-02.md))|Hariharan Devarajan (LLNL)|Suren Byna|[video][22], [pdf][23]| 22 | |11/30|Tiled: A Service for Structured Data Access ([abstract](abstracts/2021-11-30.md))|Dan Allan (NSLS-II - Brookhaven National Laboratory)|Bjoern Enders|[video][24]| 23 | |12/6 |funcX: Federated FaaS for Scientific Computing ([abstract](abstracts/2021-12-06.md))|Ryan Chard (Argonne National Laboratory)|Jonathan Skone, Bjoern Enders|[video][25]| 24 | |12/7 |Ceph Storage at CERN ([abstract](abstracts/2021-12-07.md))|Dan van der Ster, Pablo Llopis Sanmillan (CERN)|Alberto Chiusole|[video][26]| 25 | |12/14| Characterizing Machine Learning I/O Workloads on Leadership Scale HPC Systems ([abstract](abstracts/2021-12-14.md))|Ahmad Maroof Karimi (NCCS - ORNL)|Hai Ah Nam, Wahid Bhimji|[video][27]| 26 | 27 | [1]: https://drive.google.com/file/d/1oNg8YwAXeenRmyFoUNJT0I8ALol6eee8/view 28 | [2]: https://drive.google.com/file/d/1MKU_qixEq550ww4EihVin2fuoXF1QiyN/view?usp=sharing 29 | [3]: https://www.youtube.com/watch?v=vEjtb0FTS4k&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 30 | [4]: https://www.youtube.com/watch?v=LD4Zs8OCrOE&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 31 | [5]: https://drive.google.com/file/d/1mHsgiheOlD1XguNEISiWr8_ydX5XLwEk/view?usp=sharing 32 | [6]: https://www.youtube.com/watch?v=-ck2GN75ycA&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 33 | [7]: https://www.youtube.com/watch?v=YSc07PTeExw&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 34 | [8]: https://www.youtube.com/watch?v=EdvmicKME7Y&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 35 | [9]: https://docs.google.com/presentation/d/1J_xRTvOwtYeXdfqxwD5-sAswdTOw6nNo/edit?usp=drive_web&ouid=111822247062274782499&rtpof=true 36 | [10]: https://drive.google.com/file/d/18kwOhBFsGedxpee0Inb7Zh_0zOtssPG4/view?usp=sharing 37 | [11]: https://www.youtube.com/watch?v=hdcwthKcAVg&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 38 | [12]: https://www.youtube.com/watch?v=rpfun5SPFQs&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 39 | [13]: https://drive.google.com/file/d/1Zrlsv5ITqoE-ulzvqK9ok_XtZ27c5F4C/view?usp=sharing 40 | [14]: https://www.youtube.com/watch?v=Fcm0jQXqlp0&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 41 | [15]: https://docs.google.com/presentation/d/1IEd6tPPbMZOOboevNi_iYRwmPHPzdvC3XLQv0eLXsFQ 42 | [16]: https://www.youtube.com/watch?v=WRb5QDPNCrA&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 43 | [17]: https://www.youtube.com/watch?v=OQcGSaLWESE&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 44 | [18]: https://www.youtube.com/watch?v=LRkOLKww7nI&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 45 | [19]: https://www.youtube.com/watch?v=qFXJCJTSeR4&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 46 | [20]: https://www.youtube.com/watch?v=ZxboaaYoYrw&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 47 | [21]: https://www.youtube.com/watch?v=l6ygTAqmDSs&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 48 | [22]: https://www.youtube.com/watch?v=h5j8KiczIiU&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 49 | [23]: https://drive.google.com/file/d/1nVG3R2lgKg11alOGljHoLY_WuPkECXBx/view?usp=sharing 50 | [24]: https://www.youtube.com/watch?v=M5A1B0CtTg0&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 51 | [25]: https://www.youtube.com/watch?v=vMSgcjET9Uk&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 52 | [26]: https://www.youtube.com/watch?v=pkpTgGsLH_s&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 53 | [27]: https://www.youtube.com/watch?v=wFjPPkvYLhQ&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy 54 | -------------------------------------------------------------------------------- /2020.md: -------------------------------------------------------------------------------- 1 | ### 2020 Seminars 2 | |Date |Title |Speaker |Host |Material | 3 | |-----|---------------------|------------------------|-------------------|-------------| 4 | |1/10 |Independent metadata updating for large scale parallel I/O systems ([abstract](abstracts/2020-01-10.md))|Tonglin Li (NERSC) |Prabhat |[pdf][1], [vid][2]| 5 | |1/31 |Data skeletons: IO workload characterization for the modern age ([abstract](abstracts/2020-01-31.md))|Avani Wildani (Emory University)|Taylor Groves | [vid][3]| 6 | |2/07 |Time-series Analysis of ESnet Network Traffic: Statistical and Deep Learning models ([abstract](abstracts/2020-02-07.md))|Mariam Kiran (ESNet)|Steven Farrell | [vid][4]| 7 | |2/14 |Intrinsic computation and physics-based machine learning for emergentself-organization in far-from-equilibrium systems ([abstract](abstracts/2020-02-14.md))|Adam Rupe (UC Davis)|Karthik Kashinath| [vid][5]| 8 | |2/28 |The Superfacility project: 2019 year in review ([abstract](abstracts/2020-02-28.md)) | The Superfacility Project Team| Debbie Bard| [vid][9]| 9 | |3/06 |Intersections of AI/ML and Chemistry in Catalyst Design and Discovery ([abstract](abstracts/2020-03-06.md))|Zachary Ulisii (CMU)|Mustafa Mustafa| [pdf][6], [vid][7]| 10 | |3/13 |ECP HDF5 - New features and applications ([abstract](abstracts/2020-03-13.md))|Suren Byna (CRD), Quincey Koziol (NERSC)|Quincey Koziol| [pptx][8], [vid][10]| 11 | |4/17 |A Data-Driven Global Weather Model Using Reservoir Computing ([abstract](abstracts/2020-04-17.md))|Troy Arcomano (Texas A&M)|Jaideep Pathak|[vid][12]| 12 | |5/01 |Deep learning for PDEs, and scientific computing with JAX ([abstract](abstracts/2020-05-01.md))|Stephan Hoyer (Google)|Karthik Kashinath| vid coming soon| 13 | |5/15 |Deep learning production capabilities at NERSC ([abstract](abstracts/2020-05-15.md))|Steven Farrell & Mustafa Mustafa|Prabhat| [slides][11], [vid][13]| 14 | |5/22 |Learned discretizations for passive scalar advection in a 2-D turbulent flow ([abstract](abstracts/2020-05-22.md))|Jiawei Zhuang (Harvard Univ.)|Mustafa Mustafa| [slides][14]| 15 | |5/29 |Simulation-based and label-free deep learning for science ([abstract](abstracts/2020-05-29.md))|Ben Nachmann (LBL)|Wahid Bhimji| [slides][15], [vid][16]| 16 | |6/05 |Tuning Floating Point Precision (Using Dynamic and Temporal Locality Program Information) ([abstract](abstracts/2020-06-05.md))|Costin Iancu (CRD, LBL)|Brandon Cook| coming soon| 17 | |6/19 |Status of Containers in HPC ([abstract](abstracts/2020-06-19.md)) |Shane Canon |Prabhat |[vid][18]| 18 | |6/26 |Congestion and Distributed Training of Deep Neural Networks ([abstract](abstracts/2020-06-26.md)) |Jacob Balma (HPE) |Steven Farrell | [slides][17]| 19 | |7/02 |Workflows at NERSC: Overview and GNU Parallel Parsl, Papermill demos ([abstract](abstracts/2020-07-02.md)) |Bill Arndt, Laurie Stephey, Bjoern Enders|Prabhat| | 20 | |7/17 |Superconducting Radio-Frequency Cavity Fault Classification Using Machine Learning ([abstract](abstracts/2020-07-17.md)) |Christopher Tennant (Jefferson Lab)|Mustafa Mustafa|[vid][26]| 21 | |7/24 |Making the invisible visible ([abstract](abstracts/2020-07-24.md)) |Srigokul Upadhyayula|Suren Byna|[vid][21]| 22 | |8/21 |Weights & Biases: system of record to track, optimize, and reproduce ML research ([abstract](abstracts/2020-08-21.md))|Chris Van Pelt, Charles Frye (Weights & Biases)|Mustafa Mustafa| | 23 | |9/04 |Steps toward holistic control of particle accelerators with neural networks ([abstract](abstracts/2020-09-04.md))|Auralee Edelen (SLAC National Accelerator Laboratory)|Debbie Bard|[vid][25]| 24 | |9/11 |Flux: Overcoming Scheduling Challenges for Exascale Workflows ([abstract](abstracts/2020-09-11.md))|Dong Ahn, Stephen Herbein (LLNL)| Katie Antypas|| 25 | |9/25 |ExaHDF5: An Update on the ECP HDF5 Project ([abstract](abstracts/2020-09-25.md))|Quincey Koziol (NERSC), Suren Byna (CRD)|Wahid Bhimji|[slides][23], [vid][24]| 26 | |10/02 |Enabling Interactive, On-Demand High Performance Computing for Rapid Prototyping and Machine Learning ([abstract](abstracts/2020-10-02.md))|Albert Reuther (MITLincoln Laboratory Supercomputing Center)|Rollin Thomas|[vid][19], [slides][22]| 27 | |10/09 |Generative neural networks: Data-driven simulations for particle physics ([abstract](abstracts/2020-10-09.md))|Ramon Winterhalder (Heidelberg University)|Wahid Bhimji|| 28 | |10/16 |Using Machine Learning to Augment Coarse-Grid Computational Fluid Dynamics Simulations ([abstract](abstracts/2020-10-16.md))|Jaideep Pathak (NERSC)|Mustafa Mustafa|[vid][20]| 29 | |11/02 |Robust prediction of high-dimensional dynamical systems using Koopman deep networks ([abstract](abstracts/2020-11-02.md))|Omri Azencot (Ben-Gurion University)|John Wu|| 30 | 31 | [1]: https://drive.google.com/file/d/0B_vRw1QFsEicQVRuUDJpWmNFS3ZfRmNyc3pIbGFpeVdnWHZ3/view?usp=sharing 32 | [2]: https://www.youtube.com/watch?v=f2pZ6vIKCnQ&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=2 33 | [3]: https://www.youtube.com/watch?v=1lvpEwIlk_8&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=1 34 | [4]: https://www.youtube.com/watch?v=CJp_oXcgerU&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=7 35 | [5]: https://www.youtube.com/watch?v=gjx2jm25gHs&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=2&t=0s 36 | [6]: https://drive.google.com/file/d/12FrB5KhGKAMjgIprbfJqE3J5y1whuWMT/view?usp=sharing 37 | [7]: https://www.youtube.com/watch?v=cThCoWQn4-o&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=2&t=0s 38 | [8]: https://drive.google.com/file/d/1-R83QfIeZmByV-U1rqxHxFgh15WuMuJC/view 39 | [9]: https://www.youtube.com/watch?v=tcQGohF9DCg&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=3 40 | [10]: https://www.youtube.com/watch?v=VxpkNFSwpgs&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=2 41 | [11]: https://docs.google.com/presentation/d/1tXf_F2V7qaP0uDMKulsgRPUo5R3-5TM1eQa09AJ1EO0/edit?usp=sharing 42 | [12]: https://www.youtube.com/watch?v=Ujsk0dQnG1w&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=1 43 | [13]: https://www.youtube.com/watch?v=10ImerzWIkM&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=2&t=0s 44 | [14]: https://docs.google.com/presentation/d/1tFTjoJ_Ca6ABvCGBDsgba84h5YD6xve7HzzzYXn1uME/edit#slide=id.p 45 | [15]: https://www.dropbox.com/s/ei197sfm8jcs7la/NERSCMay2020.pdf?dl=0 46 | [16]: https://www.youtube.com/watch?v=1zlYYGP874I&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=2 47 | [17]: https://drive.google.com/file/d/1ITkmSVrNBrgjnbER_qh4pr4K9tkVtTnT/view?usp=sharing 48 | [18]: https://www.youtube.com/watch?v=ObXSq4fkCKs&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=3&t=5s 49 | [19]: https://www.youtube.com/watch?v=lmhaIGbz2Zo 50 | [20]: https://www.youtube.com/watch?v=2Ab-8xTI89c&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=1&t=20s 51 | [21]: https://www.youtube.com/watch?v=Ubkc1vaT1Mg&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=2&t=0s 52 | [22]: https://drive.google.com/file/d/17KGqX5CvxIDff5analOeCw0l_R4mpFcz/view?usp=sharing 53 | [23]: https://drive.google.com/file/d/1ZBJUB10IjaUpoQXyx1mP4At_ylXc-WTa/view 54 | [24]: https://www.youtube.com/watch?v=QmvAcwkTRls 55 | [25]: https://www.youtube.com/watch?v=teAJ9MOdOmE&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=4 56 | [26]: https://www.youtube.com/watch?v=7j6R0dAiydM&list=PL20S5EeApOSvkewFIuz2scAEkbnBIlzYy&index=7 57 | --------------------------------------------------------------------------------